Sie sind auf Seite 1von 7

Technical Papers

Risk management and


long-term production schedule optimization
at the LabMag iron ore deposit
in Labrador, Canada
by M. Spleit and R. Dimitrakopoulos

Abstract ■ A long-term mine production schedule optimization based on stochastic integer


programming is presented that manages the risk due to geological variability and uncertainty
while jointly considering mine profitability, truck fleet requirements and waste management.
The evolution of the pit is controlled so that space within the mined-out pit is continuously
provided to allow for tailings and waste rock to be replaced, thus minimizing the project’s
environmental footprint. The method is applied at a taconite iron ore deposit and has a project
net present value that is 16.9 percent higher than that of a conventional schedule and has a
higher chance of meeting production forecasts. The stochastic schedule reduces equipment
requirements by delaying waste mining and by mining selectively in order to meet plant ore
feed requirements without stockpiling material, resulting in 23.7 and 26.2 percent lower mining
capital and operating costs, respectively.

Mining Engineering, 2017, Vol. 69, No. 10, pp. 47-53.


Official publication of the Society for Mining, Metallurgy & Exploration Inc.
https://doi.org/10.19150/me.7807

Introduction (3.6 billion st) of proven reserves at ally being realized.


The LabMag iron ore deposit is an average total iron content of 29.8 Conventional mine planning opti-
part of the Millennium Iron Range, a percent and is very low in impurities. mization methods are based on a single
210-km (130-mile) belt of taconite in However, on a project basis of 23 Mt/a deterministic orebody model and can
northern Quebec and Labrador, Can- (25 million stpy) of product, the re- yield misleading results because they
ada. LabMag has significant economic quired capital expenditure is estimated do not account for the likely deviation
potential. It contains more than 3.3 Gt at more than $5 billion (SNC-Lavalin, from the model in reality (Ravenscroft,
2014), which necessitates careful evalu- 1992; Dowd, 1997; Dimitrakopoulos,
M. Spleit, member SME, was ation of all sources of risk. Resource es- Farrelly and Godoy, 2002; Godoy and
formerly at New Millenium Iron timation is one of the main sources of Dimitrakopoulos, 2004; Montiel, Dimi-
Corp., Montreal, Quebec, Canada, risk in a mining project as knowledge of trakopoulos and Kawahata, 2016). In
and is currently research engineer the orebody is primarily based on drill- order to consider the geological risk of
at Ecole Polytechnique de Montreal, ing, which is often sparse because it is an orebody, a set of different scenarios
Montreal, Quebec, Canada. R. expensive. If the expected ore tonnages can be created that are all equally prob-
Dimitrakopoulos, member SME, is and qualities are not obtained when able representations of the orebody
professor at COSMO – Stochastic mining, the project cash flows are di- and which all reproduce the orebody’s
Mine Planning Laboratory, rectly affected. The expected quantities spatial variability (Goovaerts, 1997;
McGill University, Montreal, and qualities of ore and waste are de- Remy, Boucher and Wu, 2009; Soares,
Quebec, Canada. Email: roussos. fined by the mine production schedule, Nunes and Azevedo, 2017). Such geo-
dimitrakopoulos@mcgill.ca. Paper which specifies the sequence of extrac- statistical simulations can be used to
No. TP-16-034. Original manuscript tion and is dependent on the resource quantify the various elements of risk
submitted August 2016. Revised estimation. The goal of mine produc- associated with a mining project: oper-
manuscript accepted for publication tion scheduling is thus to maximize the ating costs, capital costs, royalties, com-
June 2017. Discussion of this peer- expected profit, while also meeting all modity prices, taxation, tonnage and
reviewed and approved paper is production targets and constraints, by grade (Dowd, 1997; Godoy and Dimi-
invited and must be submitted to creating an extraction schedule that is trakopoulos, 2011; Montiel, Dimitrako-
SME Publications by Jan. 31, 2018. robust in the face of geological risk and poulos and Kawahata, 2016). Geologi-
has the highest chance possible of actu- cal uncertainty can then be managed
www.miningengineeringmagazine.com Mınıng engıneerıng    october 2017 47
Technical Papers
by directly incorporating stochastic simulations within the Stochastic integer programming (SIP) formulation
mine scheduling framework. The following objective function is defined as maximiza-
One flexible method for long-term production schedul- tion of the NPV minus various penalty terms that control
ing is based on stochastic integer programming (SIP) (Birge the geological risk profile, minimize fleet requirements, and
and Louveaux, 1997), a type of mathematical programming promote mining adjacent blocks in the same period in order
and modeling method that considers multiple, equally prob- to generate a practical schedule.
able scenarios and generates the optimal result for a set of
defined objectives within the feasible solution space bound- Notation
ed by a set of constraints. SIP for mine scheduling is intro- The constant and variable factors used in the SIP model
duced in Ramazan and Dimitrakopoulos (2005, 2013) and are defined below:
Dimitrakopoulos and Ramazan (2004, 2008). It proposes a
formulation that maximizes the net present value (NPV) • i, N — Index variable and total number of blocks in
while minimizing deviations from production targets using the orebody model.
a different penalty for each target. In Leite and Dimitrako- • s, S — Index variable and total number of orebody
poulos (2014), this formulation was applied at a copper de- simulations.
posit and produced a risk-robust NPV that was 29 percent • t, P — Index variable and total number of time pe-
higher than that of a conventional schedule. In Benndorf riods.
and Dimitrakopoulos (2013), an SIP formulation that inte- • d, D — Index variable and total number of destina-
grated joint multi-element geological uncertainty was ap- tions.
plied at an iron ore deposit. Additional considerations are • q, Q — Index variable and total number of metal-
easily incorporated into the modeling framework: two other lurgical qualities.
relevant studies use SIP to optimize the NPV while simul- • Vi,d,s — Value of block i in simulation s going to des-
taneously optimizing the cut-off grades (Menabde et al., tination d in time period t.
2007) and incorporating simulated future grade control data • THd,t — Total truck hours in period t for destination
at a gold deposit (Jewbali, 2006). Boland, Dumitrescu and d.
Froyland (2008) demonstrated stochastic formulations for • CtTH — Operating cost, in $/h, for trucking, discount-
mine production scheduling with endogenous uncertainty, ed by period t.
in which decisions made in later time periods can depend • bi,d,t — Binary variable with a value 1 if block i is
on observations of the geological properties of the material mined in period t and sent to destination d, and 0
mined in earlier periods. Most recently, they characterized otherwise.
the minimal sufficient constraints for such formulations so • Penconc — The penalty per tonne of deviation, in $/t
that solving them is more efficient (Boland et al., 2014). Re- concentrate, from the target concentrate production
lated extensions of stochastic formulations for the simulta- in each period; constant.
neous optimization of mineral value chains can be found in • Penq — The penalty per tonne of quality content, $/t
Montiel and Dimitrakopoulos (2017) and Goodfellow and quality q, in each period above or below the associ-
Dimitrakopoulos (2017). ated upper or lower limits, respectively; constant.
In the present work, an SIP formulation extending the • devtconc — Concentrate tonnes in excess of the upper
one in Benndorf and Dimitrakopoulos (2013) is presented limit.
to manage risk and control the risk profiles of mine produc- • devtconc — Concentrate tonnes less than the lower
tion in terms of four underlying metallurgical properties: limit.
(1) head iron, designated as FeH, (2) Davis Tube weight • devtq — Tonnes of metal or mineral, q, in excess of
recovery, as DTWR (Schulz, 1964), (3) concentrate iron, as the upper limit, where q = 1, … , Q considered quali-
FeC, and (4) silica grades, as SiC. The existing conventional ties.
schedule has two classifications of ore, with the lower-grade • devtq — Tonnes of metal or mineral, q, less than the
ore being stockpiled. The stochastic optimization seeks to lower limit.
avoid stockpiling and lower costs by deciding dynamically • ct = 1/((1 + r)t − 1) — A function for discounting prof-
to which of two destinations to send each block — the waste its and costs with the discounting factor r according
dump or the process plant — while still meeting grade con- to the period t when the block is mined.
straints. In addition to aspects included in past SIP formula- • gt = 1/((1 + GRD)t − 1) — A function for discounting
tions and applications, constraints are added to smooth the geological risk with the discounting factor GRD ac-
annual haul truck fleet requirements in order to avoid pur- cording to the period t when the block is mined.
chases that lead to underutilized equipment. Furthermore, • NRi,s — Net revenue.
and unlike past formulations, the SIP formulation presented • OCost, WCost — Mining cost of ore and waste, re-
here seeks to maximize the space available for in-pit tailings spectively, excluding truck haulage, which is penal-
disposal, which is roughly two-thirds of the ore to be mined ized directly in the objective.
at LabMag, in order to reduce the environmental footprint. • PCost — Processing cost: Crushing, concentrating,
An SIP formulation for long-term production scheduling filtration, pelletization, transportation, administra-
with equipment and tailings management is then presented. tion.
This formulation is applied in a case study at LabMag, and • ROM — Run-of-mine tonnage.
the scheduling results are compared with conventionally • eWR — Effective weight recovery: Considers ideal
scheduled results. weight recovery as well as plant efficiency parameters.

48 october 2017    Mınıng engıneerıng www.miningengineeringmagazine.com


Technical Papers
Mining block economic value. The undiscounted value added to enforce that the number of trucks in each period
for each block is defined as: must be equal to or more than the number of trucks in the
previous period. Leeway of half the available working hours
{
NRi,s − CONCi,s * PCOST − ROMi,s *
(1) of one truck in one period is given as:
Vi,d,s = OCOST − Wi,s * WCOST, if d = 1
− (ROMi,s * Wi,s  ) * WCOST, if d = 0
(4)
CONCi,s = ROMi,s * eWRi,s (2)

Objective function. The objective function of the SIP where t = 2, … , P and Hd is the total time required for trans-
model is constructed as the maximization of a profit func- portation of a block to destination d.
tion, defined as the total expected NPV minus penalties
for deviations from planned production targets and pen- Application at the LabMag iron ore deposit
alties for not mining the blocks adjacent to a mined block The formulation is applied at the LabMag deposit in
(Ramazan and Dimitrakopoulos, 2005, 2013; Benndorf and order to create a mine production schedule that considers
Dimitrakopoulos, 2013), as well as truck operating costs. multi-element grade uncertainty as well as equipment and
This objective function includes four distinct terms: tailings disposal requirements.
Maximize Obj = Stochastic orebody models at LabMag. Geological vari-
ability and uncertainty is represented here by 10 stochastic
(3a)
conditional simulations of the deposit (Spleit and Dimitrako-
poulos, 2016). Each realization consists of a joint simulation
(3b) of the seven correlated layer thicknesses as well as the joint
simulation of four correlated ore characteristics in each layer
using the joint direct block simulation method (Boucher and
(3c) Dimitrakopoulos, 2009). Each model consists of 13,400 blocks,
measuring 100 × 100 × 15 m (328 × 328 × 49 ft).
The two primary waste types for the LabMag deposit are
(3d) overburden (OB) and Menihek Shale (MS). The OB over-
lies the entire deposit and is minimal (the underlying rock
The first term is the primary term and represents the NPV of is commonly exposed at surface). The MS layer is present
all blocks mined in the optimization. The second term rep- on the northeast side of the deposit, overlies the iron layers
resents the trucking operating cost, which is minimized. The and dips parallel to them at approximately 6 degrees (Fig. 1).
third term acts to penalize deviations from target concen-
trate tonnes and the target silica grade and weight recovery. Implementation. The SIP model was implemented in Vi-
The variables for the deviations are determined by the op- sual C++ using the ILOG CPLEX API. The initial attempts
timization process, based on the corresponding constraints to solve the full orebody model for all 10 periods proved
that are set. The fourth term is a penalty for not mining that it was unsolvable in a reasonable amount of time, as the
adjacent blocks. It is desirable to mine blocks in groups in optimization had made little progress after several days of
order to generate a practical schedule. There is a trade-off running on a 64-bit Dell Precision M6500 Intel I7 quad-core
between the penalty terms, determined by the relative size at 1.73 GHz with 16 GB of RAM. The initial 13,400 blocks
of the penalties. considered are the blocks contained within an ultimate pit
derived using the nested Lerchs-Grossman algorithm. As
Constraints. Constraints are applied to ensure equip- only the first 10 years are scheduled within the optimization
ment accessibility, processing capacity, geotechnical aspects, and the LabMag ultimate pit contains more than 25 years
and blending requirements, which are detailed in Benndorf of ore at the planned capacity, the precise pit limit need not
and Dimitrakopoulos (2013). In addition, constraints are in- be discussed further. To reduce the number of blocks, a new
cluded here to control the schedule sequence and the num- pit was designed that takes as many blocks as possible but
ber of haul trucks in each period. avoids the MS waste layer. As the optimization targets the
first 10 years and tries to minimize trucking hours as well
Sequencing constraints. In order to accommodate in-pit
tailings disposal, the strategy here is to expose the ultimate
pit floor as quickly as possible. In order to accomplish this, Figure 1
mining on each bench is set to start from the shallow side of LabMag typical cross-section.
the deposit. In addition to the standard block predecessors
needed to maintain geotechnical slope constraint, the block
cross-dip, toward where the deposit daylights at surface, is
set to be mined in the same or an earlier period.

Haulage capacity constraints. To avoid a truck being


leased or purchased and then left unused, a constraint is

www.miningengineeringmagazine.com Mınıng engıneerıng    october 2017 49


Technical Papers
as unnecessary waste, it was evident that the optimization mining width, as well as the same targets and constraints. In
would avoid the MS region of the deposit, anyway. This the stochastic optimization, the only blocks sent to the waste
brought the number of blocks down to 8,411. The optimi- dump are located at the surface of the deposit and contain
zation was then broken down into four sequential optimi- mostly OB and/or MS waste, which means that all scheduled
zations (Fig. 2), each set to stop once there was less than a material within the seven iron-bearing units is sent to the
1 percent gap between the solution and the optimal solution. plant. Had ore blocks with a low DTWR been sent to the
Note that the four sequential optimizations do not have an dump, this would have promoted the concept of stockpiling
impact on the validity of the comparisons made later in this ore with a low weight recovery. However, this is not the case,
paper between the resulting stochastic schedule and the con- so the only reason to stockpile ore would be to restrict the
ventional one. If anything, solving the proposed formulation silica levels. In the optimized schedule, the silica levels are
in a single optimization step, instead of as four subproblems, managed without the need for such a restriction.
will only improve the value of the objective function and The evolution of the pit in the stochastic optimization
thus be in favor of the stochastic solution. schedule is along the full length of the deposit, progressively
The optimization has upper and lower bounds of 23 and deepening perpendicular to the strike (Fig. 3). This differs
29 percent for DTWR and 1.8 and 2.5 percent for SiC, and from the conventional design that is based on slots that are
penalizes values outside those bounds. Higher DTWR means mined west to east along the full width of the defined re-
mining less ore for the same amount of product, but there is a source and mining at depth already in early periods (Fig. 4).
trade-off with haul distance because most of the higher-grade The stochastic optimization schedule thus has shorter haul
DTWR material is located in the north end of the deposit, times in earlier periods and fewer trucks as they can travel
which is further from the crusher. The average silica grade of at near top speed, whereas they are driven slower when ex-
the single, 10-year period of optimization is 2.2 percent, which iting the pit due to the incline. Another advantage of min-
became the new target for subsequent optimizations because ing along the length of the deposit is that the grades vary
a consistent silica blend is desired across all periods. primarily along the strike: higher DTWR material is found
to the north but with higher SiC as well. Having open faces
Results along the full length of the deposit allows for more flexibility
A practical mining schedule was designed based on the during operations to achieve the necessary blend.
block-scale optimization. The optimization considers the Figure 5 shows that the truck productivity of the opti-
first 10 years, and an additional 15 years were scheduled mized schedule is at its maximum in the earlier periods, and
thereafter to allow for full comparison against an existing steadily declines with each period as the pit is deepened and
conventional schedule. Both conventional and stochastic the cycle times increase. For the deterministic schedule, the
mine designs and production schedules are based on the productivity seesaws as each full-width slot is mined. The
same parameters, such as pit slopes, bench height and safe optimized schedule ensures higher productivity in ear-
lier periods and thus lower corresponding operating costs.

Figure 2 Risk management. The stochastic optimization sched-


Schematic representation of four-stage schedule ule is compared with the conventional schedule in order to
optimization. demonstrate the effectiveness of geological risk manage-
ment. The annual concentrate product tonnages for a con-
ventional schedule, designed based on a single deterministic
orebody model, and a stochastic schedule, which considers
simultaneously several different and equally probable ore-
body models along with the local spatial variability of perti-
nent geological attributes, are shown in Figs. 6a and 6b. The
expected tonnages for the stochastic schedule are consistent
with the targets and also have a greatly reduced (narrow)
risk profile, which is shown by a tight 80 percent confidence

Figure 3 Figure 4
Stochastic schedule mine designs: end-of-period (a) 5, (b) 10 Deterministic schedule mine designs: end-of-period (a) 5,
and (c) 15. (b) 10 and (c) 15.

(a) (b) (c) (a) (b) (c)

50 october 2017    Mınıng engıneerıng www.miningengineeringmagazine.com


Technical Papers

Figure 5 Figure 6
Conventional and stochastic schedules: Comparison of truck Product tonnes in (a) conventional schedule and (b)
productivity. stochastic schedule.
(a)

range. The conventional schedule considers a single tonnage (b)


estimate for each block, essentially treating that information
as 100 percent accurate, which is not true. The actual tonnes
mined from each block may end up being lower or higher
than the estimate (an average value by construction), and
when these differences are taken together they can trans-
late to a large deviation from the expected production over a
mining period. The stochastic optimization jointly considers
the probabilistic distribution of possible values for each in-
dividual block and schedules block combinations that have a
higher probability of meeting production targets.
For the conventional schedule in Fig. 6a, the solid black
line shows the target tonnage of 22 Mt/a (24 million stpy) with
a ramp-up, and the grey points show the forecasted values
Figure 7
when evaluating the performance of the conventional sched- (a) Annual DTWR of feed ore and risk profile in the stochastic
ule using the deterministic (estimated) orebody model. The schedule and (b) annual silica grades of concentrate and risk
conventional schedule was designed based on the estimated profile in the stochastic schedule.
orebody model and, in general, Fig. 6a shows that tonnages
are reasonably consistent with the targets. Tonnages in excess (a)
of the target are due to including inferred material as pro-
cessed ore, and demonstrates the possibility of mining more
ore than necessary. However, these single estimated values
and forecasts generated by the conventional schedule are mis-
leading. This is shown by the red bounds that quantify the risk
in meeting the conventional forecasts through a range of pos-
sible tonnages for each period and with 80 percent confidence
level that the true value will be in this range. The tonnage
ranges are obtained by evaluating the conventional schedule
against each of orebody simulations (equally likely represen- (b)
tations of the actual deposit). Note that the dashed light blue
line in Fig. 6a shows the expected value for each production
year, calculated as the average from the set of values obtained.
The probabilistic evaluation (risk analysis) shows that the de-
terministic model systematically overestimates the quantity
of concentrate tonnes. There are two reasons for this overes-
timation: (1) the deterministic model does not assume any lo-
cal variability and has smoothed out estimates of the weight
recovery, and (2) to account for uncertainty, the simulations
have different values of FeH and the density of each lithology
is dependent, and calculated using regressions, on the FeH in
each layer, whereas the deterministic model assumes a fixed 7b to demonstrate the effectiveness of the stochastic schedule
average density for each lithology. to control the qualities and their risk profiles. The DTWR val-
The annual values for the stochastic schedule for the two ues vary by only 0.4 percent on average. High DTWR values
primary qualities, DTWR and SiC, are shown in Figs. 7a and in earlier periods were expected because a greater weight re-

www.miningengineeringmagazine.com Mınıng engıneerıng    october 2017 51


Technical Papers
covery means less ore is required for the same product ton- maximizes the discounted cash flows. This can be accom-
nage, which means lower costs. That is not the case here be- plished by minimizing waste mining, minimizing haulage
cause the optimization seeks the greatest profit, and there is a cycle times and scheduling higher-grade material. The sto-
balance between the value of higher DTWR material and the chastic optimization not only maximizes the discounted cash
higher cost of mining at depth and/or further from the plant. flows, but also ensures a greater certainty in financial fore-
However, this may be an artificial result: processing costs and casts by managing the geological variability and uncertainty.
plant efficiencies are variable with respect to feed material, Compared with the stochastic optimization schedule,
yet they are assumed as fixed in this study. With the inclusion there is a significant amount of MS waste mined in the con-
of more detailed variable processing costs and efficiencies, it ventional schedule as well as additional ore that is stockpiled
is likely that higher DTWR material would be scheduled in instead of being sent to the plant. Mining these two materi-
earlier periods before the process plant is fully commissioned als requires a significant amount of additional equipment,
and operating consistently. The SiC range for each year in which translates to higher costs than those for the stochastic
the stochastic schedule (Fig. 7b) can be seen to fall within the schedule. The expected annual waste and stockpile tonnages,
specified upper and lower limits of 2.5 and 1.8 percent, respec- based on evaluating the schedule with the simulations, are
tively, hovering around the target of 2.2 percent and demon- shown in Fig. 8a for both conventional and stochastic sched-
strating the ability of the stochastic modeling to ensure low ules. Low-grade ore is stockpiled in the conventional sched-
risk of not meeting targets. ule based on a fixed cut-off in order to ensure the desired
ore qualities going to the plant. There is no stockpile in the
Financial impact of stochastic scheduling. In addition optimization schedule, which seeks instead to meet the de-
to controlling the qualities and tonnages, and thus minimiz- sired ore quality constraints by sending material only to ei-
ing the deviations from targets, the stochastic optimization ther the process plant or the waste dump. Figure 8a shows
that the combined tonnage of waste and stockpiled ores is
roughly between 20 and 50 Mt/a (22 to 55 million stpy) for
Figure 8 all periods, whereas the stochastic schedule has no stock-
piled tonnes and only minimal waste. The stochastic optimi-
Conventional and stochastic schedules: Comparisons of (a) zation was performed on a pit that purposely avoided the
expected waste and stockpile tonnes, (b) major equipment MS, and so there is minimal stripping in all years. The very
fleets and (c) cumulative discounted cash flows (DCF). low amount of waste mining was intended, and is a crucial
component to minimizing costs in the first 10 years.
(a)
Figure 8b shows a comparison between the fleet require-
ments of the two schedules for haul trucks and primary cable
shovels. Equipment calculations take into account a variety
of factors, including mechanical availability, utilization, job
efficiency, operating delays, payload, spot times, dump times,
load times and cycle times. The new maximum number of
trucks required over the 10-year period is 20 trucks, as op-
posed to the previous 35 trucks. Fewer trucks are needed be-
cause the haul cycle times are shorter, so the trucks are more
productive. In addition, there is less waste mining, which
also contributes to the reduced number of equipment. The
(b) necessary cable shovels was reduced by one, which is signifi-
cant because each cable shovel costs almost four times as
much as one truck. The change in fleet requirements reduces
the capital cost requirements by 23.7 percent and operating
costs by 26.2 percent.
The impact of these cost reductions and the reduction
of geological variability can be seen together in a compari-
son of discounted cash flows for the two schedules (Fig. 8c).
The dashed black line shows the estimated results from the
conventional schedule when it is evaluated using a single de-
terministic orebody model. All other results are shown rela-
tive to this. The pink zone shows the 80 percent confidence
range of the conventional schedule when it is evaluated with
the stochastic simulations. Almost all simulations result in a
lower NPV and the expected value of the simulations is 14.9
percent less than the estimated value. This is directly related
to the overestimation of product tonnes by the deterministic
orebody model. The 80 percent confidence range spans 28.9
percent of reference conventionally estimated NPV, which
(c) indicates significant uncertainty in the NPV.
The stochastic optimization schedule, however, has an

52 october 2017    Mınıng engıneerıng www.miningengineeringmagazine.com


Technical Papers
expected NPV that is 16.9 percent higher than the reference References
conventionally estimated NPV. The 80 percent confidence Benndorf, J., and Dimitrakopoulos, R., 2013, “Stochastic long-term production scheduling
of iron ore deposits: integrating joint multi-element geological uncertainty,”
range is 2.5 percent, ranging from 15.6 to 18.1 percent more Journal of Mining Science, Vol. 49, No. 1, pp. 68-81, https://doi.org/10.1134/
than the NPV of the deterministic schedule, which represents s1062739149010097.
a significantly reduced risk profile. This indicates that a higher Birge, J.R., and Louveaux, F., 1997, Introduction to Stochastic Programming, Springer
Series in Operations Research, Springer, New York, 421 pp.
NPV is achieved with the stochastic schedule and there is a Boland, N., Dumitrescu, I., and Froyland, G., 2008, “A multistage stochastic
much greater probability of actually achieving this result. programming approach to open pit mine production scheduling with uncertain
geology,” Optimization Online.
Conclusions Boland, N., Dumitrescu, I., Froyland, G., and Kalinowski, T., 2014, “Minimum cardinality
non-anticipativity constraint sets for multistage stochastic programming with
A new formulation for long-term mine production endogenous observation of uncertainty,” University of New South Wales.
schedule optimization based on stochastic integer program- Boucher, A., and Dimitrakopoulos, R., 2009, "Block-support simulation of multiple
ming was applied to the LabMag iron ore deposit in order correlated variables," Mathematical Geosciences, Vol. 41, No. 2, pp. 215-273,
https://doi.org/10.1007/s11004-008-9178-0.
to maximize the mine profitability and reduce the risk of not Dimitrakopoulos, R., Farrelly, C., and Godoy, M., 2002, “Moving forward from
achieving the expected financial results by managing geo- traditional optimisation: grade uncertainty and risk effects in open pit mine design,”
logical uncertainty. Because past experience has shown that Transactions of the IMM, Section A Mining Technology, pp. A82-A89, https://doi.
org/10.1179/mnt.2002.111.1.82.
actual mining production never corresponds perfectly with Dimitrakopoulos, R., and Ramazan, S., 2004, “Uncertainty based production scheduling
forecasts, this formulation considers multiple orebody simu- in open pit mining,” Transactions of the Society for Mining, Metallurgy & Exploration,
lations that capture the range of possible ore qualities and SME, Englewood, CO, Vol. 316, pp.106-112.
Dimitrakopoulos, R., and Ramazan, S., 2008, “Stochastic integer programming for
quantities for each block and assesses the impact of differ- optimizing long term production schedules of open pit mines: methods, application
ent block groupings. Furthermore, the formulation ensures and value of stochastic solutions,” IMM Transactions, Mining Technology, Vol. 117,
numerous standard mining constraints are respected, as well No. 4, pp.155-160, https://doi.org/10.1179/174328609x417279.
Dowd, P.A., 1997, “Risk in minerals projects: analysis, perception and management,”
as two new constraints that could be relevant for many other Transactions of the IMM, Section A Mining Technology, pp. A9-18.
mining projects: one constraint restricts changes to annual Godoy, M., and Dimitrakopoulos, R., 2004, “Managing risk and waste mining in long-
equipment requirements to prevent dips that would leave term production scheduling,” Transactions of the Society for Mining, Metallurgy
equipment unused or else require a mixed fleet of owned & Exploration, SME, Englewood, CO, Vol. 316, pp. 43-50.
Godoy, M., and Dimitrakopoulos, R., 2011, “A risk quantification framework for
versus contracted equipment, while the other restricts the strategic risk management in mine design: method and application at a gold mine,”
mining progression with the goal of maximizing the volume Journal of Mining Science, Vol. 84, No. 2, pp. 235-246, https://doi.org/10.1134/
of space available in-pit for tailings and/or waste disposal. s1062739147020109.
Goodfellow, R., and Dimitrakopoulos, R., 2017, “Simultaneous stochastic optimization
The optimized schedule yielded an expected NPV that is of mining complexes and mineral value chains,” Mathematical Geosciences, https://
16.9 percent higher than that of a conventional schedule and doi.org/10.1007/s11004-017-9680-3.
has a higher chance of being realized due to the reduced un- Goovaerts, P., 1997, Geostatistics for Natural Resources Evaluation, Oxford University
Press, New York, 483 pp.
certainty in the expected concentrate product tonnages. The Jewbali, A., 2006, “Modeling Geological Uncertainty for Stochastic Short-Term
optimized schedule ensures annual DTWR values are likely Production Scheduling in Open Pit Metal Mines,” Ph.D. Thesis, School of
to vary by only 0.4 percent on average, and that the expected Engineering, University of Queensland, Brisbane, Queensland, Australia.
Leite, A., and Dimitrakopoulos, R., 2014, “Stochastic optimization of mine production
annual silica is near 2.2 percent in all years with the range scheduling with uncertain ore/metal/waste supply,” Journal of Mining Science and
of probable values all falling within the specified acceptable Technology, Vol. 24, No. 6, pp. 755-762, https://doi.org/10.1016/j.ijmst.2014.10.004.
upper and lower limits of 2.5 and 1.8 percent. By effectively Menabde, M., Froyland, G., Stone, P., andYeates, G., 2007, “Mining schedule optimization
blending ore to meet production targets, stockpiling is elimi- for conditionally simulated orebodies,” Orebody Modeling and Strategic Mine
Planning: Uncertainty and Risk Management Models, AusIMM Spectrum Series
nated from this schedule. Waste mining is reduced for all pe- 14, 2nd Edition, pp. 379-384.
riods by completely avoiding an overlying waste unit that was Montiel, L., and Dimitrakopoulos, R., 2017, “An extended stochastic optimization
mined in the previous schedule and the combined effect of no method for multi-process mining complexes,” Journal of Heuristics, https://
doi.10.1007/s10732-017-9349-6.
stockpile and very little waste mining reduced the total exca- Montiel, L., Dimitrakopoulos, R., and Kawahata, K., 2016, “Globally optimising
vation by 20 to 50 Mt/a (22 to 55 million stpy) in each year. open-pit and underground mining operations under geological uncertainty,”
Compared with the first 10 years of the previous life-of-mine Mining Technology, Vol. 125, No. 1, pp. 2-14, https://doi.org/10.1179/17432863
15y.0000000027.
schedule, the optimized schedule reduces the required num- Ramazan, S., and Dimitrakopoulos R., 2005, “Stochastic optimization of long term
ber of trucks by 15, from a previous total of 35, to 20 trucks in production scheduling for open pit mines with a new integer programming
total, and the required number of shovels by one, to five shov- formulation,” Orebody Modeling and Strategic Mine Planning, AusIMM Spectrum
Series 14, pp. 385-392.
els in total. This corresponds to reductions of 23.7 percent in Ramazan, S., and Dimitrakopoulos R., 2013, “Production scheduling with uncertain
capital costs and 26.2 percent in operating costs over the first supply: A new solution to the open pit mining problem,” Optimization and
10 years. The optimized schedule mines the orebody in a pro- Engineering, Vol. 14, No. 2, pp. 361-380, https://doi.org/10.1007/s11081-012-9186-2.
gressively deepening fashion, maintaining a larger working Ravenscroft, P.J., 1992, “Risk analysis for mine scheduling by conditional simulation,”
Transactions of the IMM, Section A Mining Technology, Vol. 101, pp. A104-108.
area at any given time, rather than mining a slot that reaches Remy, N., Boucher, A., and Wu, J., 2009, Applied Geostatistics with SGeMS, Oxford
the full depth of the deposit. This also facilitates the eventual University Press, New York, https://doi.org/10.1017/cbo9781139150019.
disposal of dry tailings and waste inside the pit in order to Schulz, N.F., 1964, “Determination of the magnetic separation characteristics with
the Davis Magnetic Tube,” Transactions of the Metallurgical Society of AIME,
reduce the environmental footprint. ■ Vol. 229, pp. 211-216.
SNC-Lavalin, 2014, “Technical Report on the Feasibility Study of the LabMag Iron Ore
Acknowledgments Deposit, Labrador, Canada,” Feasibility Study Report, New Millennium Iron Corp.
Soares, A., Nunes, R., and Azevedo, L., 2017, “Integration of uncertain data in
We gratefully acknowledge the support of New Millen- geostatistical modelling,” Mathematical Geosciences, Vol. 49, No. 2, pp. 253-273.
nium Iron Corp. and the National Sciences and Engineer- Spleit, M., and Dimitrakopoulos, R., 2016, “Modeling geological variability in the LabMag
ing Research Council of Canada (NSERC Discovery Grant iron ore deposit and effects on the long-term production schedule,” Transactions
of the IMM, Section A Mining Technology, Vol. 126, No.1, pp. 1-15, https://doi.or
239019-06). g/10.1080/14749009.2016.1235247.

www.miningengineeringmagazine.com Mınıng engıneerıng    october 2017 53

Das könnte Ihnen auch gefallen