Production Scheduling Optimisation in a Nickel Laterite Deposit:

MIP and LP Applications and Infeasibility in the Presence of

Orebody Variability
S Ramazan1 and R Dimitrakopoulos2

ABSTRACT of the actual orebody model to be mined. MIP applications in

complex multi-element deposits, such as Ni laterites or iron ore,
Optimisation and mathematical formulations for open pit design and
long-term production scheduling have been available since the 1960s.
are generally not available.
There are two main concerns in the application of existing mixed integer The effects of orebody uncertainty and in situ variability in
programming (MIP) models in open pit mines: a. the feasibility of optimising open pit mine design are shown in Dimitrakopoulos et
generating both optimal solutions and practically mineable scheduling al (2002), where the substantial conceptual and economic
patterns and b. the inability to deal with in situ variability in orebodies differences of risk-based frameworks are shown. The need to
and account for uncertainty. In dealing with these issues, a general open consider risk as well as in situ variability in production
pit production scheduling method for multi-element deposits is presented scheduling is also raised in Ravenscroft (1992). Similarly, Smith
and subsequently used to optimise the long-term production schedule in a and Dimitrakopoulos (1999) show the risk quantification in
nickel laterite deposit. The application shows the weaknesses of MIP traditional short-term production scheduling, using MIP for a
formulations in terms of generating optimal solutions as well as limited number of binary variables.
practically feasible mining patterns. The application of the same
formulation in the presence of in situ orebody variability and uncertainty The paper first develops a general MIP formulation for
uncovers a major reason for deviations of scheduled production from optimising the scheduling of long-term production of
actual production, and outlines additional limits of traditional multi-element deposits. Subsequently, the formulation is applied
optimisation formulations as well as future needs. to a Ni-Co laterite deposit. The paper discusses issues of
infeasibility for MIP models, and issues related to the limitations
of MIP formulations with respect to practical scheduling, and
INTRODUCTION accommodating equipment access and mobility. The issue of
Optimisation of the long-term production scheduling mechanism orebody uncertainty and in situ variability is explored. The MIP
is a major step in mine planning. It aims to maximise the overall formulation is applied to stochastically simulated representations
discounted net revenue to be generated from a mining project of the deposit. Conclusions and suggestions follow.
within the operational constraints, such as mining slope, grade
blending, ore production and mining capacity. MIP FORMULATIONS IN LONG-TERM OPEN PIT
Mixed integer programming (MIP) and linear programming PRODUCTION SCHEDULING
(LP) types of mathematical models are common tools in
optimising mine production scheduling. There have been several In open pit long-term production scheduling, generally, MIP
efforts in developing suitable mathematical models, including the models are constructed to maximise the overall discounted
early efforts of Johnson (1968) who introduced LP to optimising economic value, or net present value (NPV), of a mining project.
mine planning. Gershon (1983) discusses an LP approach The general MIP form of an open pit production scheduling
together with MIP models for optimising mine scheduling that problem can be represented as follows:
allows partial block mining provided that the entire block
preceding a block has been mined. Notable also is the work by The objective function
Dagdelen (1985), which formulated long-term production p n
scheduling as an MIP mathematical model and applied the
Maximise ∑ ∑c t
* x it
lagrangian and subgradient methods to solve the formulations, t=1 i=1
i

and by Akaike (1999), which used an MIP model to optimise

long-term production scheduling. where:
The key problem in these MIP models has been shown to be p is the maximum number of scheduling periods
the difficulty of solving large MIP models that require too many
binary (0-1) variables. The attempted LP-based models often n is the total number of blocks to be scheduled
generate fractional mining of the blocks, and the design becomes ci t
is the NPV to be generated by mining block i in period t
infeasible and/or not optimal when blocks are mined as a whole.
Ramazan (2001) and Johnson et al (2002) propose a new method xit is a binary variable, equal to 1 if the block i is to be mined
based on the fundamental tree concept to decrease the number of in period t, 0 otherwise
binary variables required in MIP formulations for long-term
production scheduling. The method significantly decreases the Grade blending constraints
number of binary variables required to generate an MIP model
and enables MIP models to be applied to large mine deposits. 1. Upper bound constraints: the average grade of the material
However, orebody uncertainty is not considered in any of the sent to the mill has to be less than or equal to a certain
above methods. All inputs are taken ‘as is’. In addition, the grade value, Gmax, for each period, t.
formulations are traditionally applied to smooth representations n

∑ (g
i=1
i − Gmax ) * Oi * x it ≤ 0
1. WH Bryan Mining Geology Research Centre, The University of
Queensland, Brisbane Qld 4072. where:
2. MAusIMM, WH Bryan Mining Geology Research Centre, The gi is the average grade of block i
University of Queensland, Brisbane Qld 4072. Oi is the ore tonnage in block i
E-mail: [email protected]

Mine Planning and Equipment Selection Kalgoorlie, WA, 23 - 25 April 2003 3

S RAMAZAN and R DIMITRAKOPOULOS

2. Lower bound constraints: the average grade of the material 2. Using m- constraints for each block per period:
sent to the mill has to be greater than or equal to a certain t
value, Gmin, for each period, t. x kt − ∑ x 1r ≤ 0, t = 1, 2, 3,... p
n r=1

∑ (g
i=1
i − Gmin ) * Oi * x it ≥ 0
it should be noted that the MIP formulations presented do
not consider the smoothness of the scheduled patterns,
which relates to equipment movement in a period and
Reserve constraints equipment access. The geological uncertainty is also
Reserve constraints are constructed for each of the blocks to state ignored in this traditional optimisation model.
that all the blocks in the model considered have to be mined
once. APPLICATION OF TRADITIONAL MIP SCHEDULE
p IN A NI LATERITE DEPOSIT
∑x
t=1
t
i =1
The mine data set includes Ni, Co, Mg and Al grades, volume of
percent rock and thickness of two different layers. The main
Generally, the orebody model contains many blocks and it is minerals considered for profit in this project is Ni. Co mineral is
very difficult, or impossible, to generate a solution through MIP
also a commodity sold but its contribution to overall mine profit
formulations if they are applied to the whole orebody model.
is comparatively small. Too high or too little contents of Mg and
Therefore, it is often necessary to consider only applying the
formulations to the blocks within the ultimate pit limits. Al may increase the processing cost significantly since they are
related to the acid consumption. The data set contains 2030
(58 × 35) blocks with 40 m × 40 m dimensions along north outh
Processing capacity constraints and east est directions, with variable thickness depending on the
1. Upper bound: the total tons of ore processed cannot be availability of ore. Each block may contain both ore and waste
more than the processing capacity (PCmax) in any period, t. material. The deposit is estimated to contain around 28.9 million
n tons of ore with an average grade of about 1.3 per cent Ni, 4.50
∑ (O * x ) ≤ PC
i=1
i
t
i max
per cent Mg, 0.58 per cent Al and a total of approximately 47.4
million tons of material. A production schedule for the deposit is
developed to maximise the total NPV at eight per cent discount
2. Lower bound: the total tons of ore processed cannot be less
rate, meeting the production targets in terms of periodical ore
than a certain amount (PCmin) each period, t.
production with certain grades of Ni, Al and Mg for three years
n
of mine life. The binding constraints implemented in the
∑ (O * x ) ≥ PC
i=1
i
t
i min optimisation process are provided in Table 1.

Mining capacity
TABLE 1
The total amount of material (waste and ore) to be mined cannot The constraints binding the scheduling model.
be more than the total available equipment capacity (MCmax) for
each period, t. Min Max
n Ore tons (106) 9.5 10.0
∑ (O + W ) * x
i=1
i i
t
i ≤ MC max Ni (%) 1.2 1.4
Al (%) 0.5 0.7
where: Mg (%) 4.0 5.0
Wi is the tonnage of waste material within block I
A lower bound may need to be implemented if an objective is Two types of orebody models are created in this case study.
to force the MIP model to produce balanced waste production The first is a traditional one, containing estimated values of all
throughout the periods as follows: orebody attributes of interest, modelled with ordinary kriging.
n This model is referred to herein as the ‘kriging model’. The
∑ (O + W ) * x
i=1
i i
t
i ≥ MC min second type of orebody model includes stochastically simulated
ones (Dimitrakopoulos, 1998; 2002). Ni grade and ore tonnage
distributions for the kriging model are shown in Figure 1. The
Slope constraints kriging model will be used as input in this section, as mostly the
All the overlying blocks that must be mined before mining a case in traditional optimisation methods. Simulated models are
given block have to be determined. This can be implemented used in a subsequent section.
through one or more cone templates representing the required The traditional MIP scheduling model is applied to maximise
wall slopes of the open pit mine. There are two ways of the total NPV of the project using the kriging orebody model as
implementing these constraints. input and defining all the variables for all periods as binary (0-1).
1. Using one constraint for each block per period: This gave a total of 6090 binary variables, and no feasible
m t
solution could be generated for this scheduling model. Then, the
mx kt − ∑ ∑ x 1r ≤ 0, t = 1, 2, 3,... p variables representing the last period are redefined as linear,
l=1 r=1 reducing the number of binary variables down to 4060. But still a
feasible solution could not be obtained.
where:
The scheduling model is solved in less than one minute when
K is the index of a block considered for excavation in all the variables in the model are defined as linear. To prevent
period t partial mining of the blocks, the economic value of the blocks
M is the total number of blocks overlying block k have been changed by 0.001 unit for the blocks that have the

4 Kalgoorlie, WA, 23 - 25 April 2003 Mine Planning and Equipment Selection

 PRODUCTION SCHEDULING OPTIMISATION IN A NICKEL LATERITE DEPOSIT

same economic values. For example if there are three blocks Grades (%)
having -10 value, they are changed to -10.0, -9.999, -9.998, 0.0 – 1.2
which are mathematically different, but not significantly 1.2 – 1.3
different in terms of economic considerations of the project. The 1.3 – 1.4
model produced seven blocks with partial mining and high 1.4 – 3.5
percentages of each of these blocks are scheduled in a single
period. Figure 2a shows the plan view of the schedule produced Ore Tons
0 – 12,000
by the LP model and Table 2 summarises the results, showing
12,000 – 13,500
that the LP model constraints are respected. However, since this 13,500 – 14,500
LP model does not contain any smoothness factor, the resultant 14,500 – 30,000
schedule is widely spread in individual periods. This wide spread
of the scheduling pattern raises questions on the feasibility and 500m
optimality of this schedule. It requires a lot of equipment
movement in a single period, which the model did not take in to
account for optimisation, and some of the blocks are difficult to
excavate because there is not enough space around them for
equipment access. FIG 1 - Ni grade (left) and ore tonnage (right) distributions of
Because the scheduling pattern obtained from the LP model is kriging model.
not a practically feasible mining pattern, it may be useful to
consider smoothing the scheduling images according to their
location and the scheduled periods given in Figure 2a. Figure 2b
shows the image after smoothing and Table 3 summarises the
results of the smoothed schedule. The feasibility of the Periods
scheduling pattern in terms of required equipment mobility in a 1
single period may be still questionable, but it is largely improved 2
3
over the original MIP output. In addition to the implementation
issues, the table shows that in the first year ore production
constraint and in the second year the ore production and Al per
cent constraints are violated by smoothing the schedule.

ANALYSIS OF THE SCHEDULE CONSIDERING

OREBODY UNCERTAINTY
In this section, the jointly simulated models of the Ni deposit are 500m
used. Figure 3 shows the Ni grade distribution of the five
simulated orebody models and Figure 4 shows the ore tonnage (a) (b)
distributions, which, although similar in general, have significant
local differences. These five orebody models are scheduled using
the same MIP scheduling model used in the traditional
optimisation. Figure 5 presents the scheduling patterns of these FIG 2 - Plan view of the schedule obtained using kriging model.
models as direct output of the MIP model (left) and after (a) is the output of LP model; (b) is the smoothed image of (a).
smoothing the image (right).

TABLE 2
Summary results from the LP schedule of kriging model. T. tons is the total tonnage; O. tons is the ore tonnage; UEV is economic
undiscounted value; S./Ave is the sum of the columns for tonnage and economic values and average of the columns representing grades.

Time (years) T. Tons (106) O. Tons (106) UEV ($106) NPV ($106) Ni (%) Co (%) Mg (%) Al (%)
1 14.94 9.83 719.27 635.12 1.400 0.109 4.242 0.700
2 15.69 9.50 563.20 459.82 1.295 0.084 4.927 0.700
3 17.69 9.50 372.78 343.71 1.200 0.071 5.000 0.629
S./Ave 48.32 28.83 1655.2 1438.66 1.300 0.088 4.718 0.676

TABLE 3
Summary results after smoothing the LP schedule of kriging model. T. tons is the total tonnage; O. tons is the ore tonnage; UEV is economic
undiscounted value; S./Ave is the sum of the columns for tonnage and economic values and average of the columns representing grades.

Time (years) T. Tons (106) O. Tons (106) UEV ($106) NPV ($106) Ni (%) Co (%) Mg (%) Al (%)
1 16.70 10.88 736.66 682.09 1.384 0.105 4.320 0.683
2 14.06 8.51 479.74 411.30 1.292 0.084 4.886 0.731
3 17.56 9.44 438.85 348.37 1.210 0.072 5.023 0.621
S./Ave 48.32 28.83 1655.2 1441.77 1.300 0.088 4.718 0.676

Mine Planning and Equipment Selection Kalgoorlie, WA, 23 - 25 April 2003 5

S RAMAZAN and R DIMITRAKOPOULOS

significant cost of the required equipment movements in the

optimisation, and many blocks do not have sufficient equipment
access. The scheduling patterns are smoothed by considering the
location of the blocks and scheduling periods of the surrounding
blocks to improve their practicality. Due to the blocks being
scheduled too widely over the mineral deposit in a period, there
are substantial differences between smoothed images and the
MIP output. Changing the scheduling periods of many blocks
often causes the schedule to be infeasible for the considered
operational requirements such as ore tons and grades. And if the
schedule is still feasible, it is probably far from being optimal.
Clearly, the application of traditional optimisation formulations
Grades (%) for production scheduling is weak in the presence of in situ
0.0 – 1.2 variability and uncertainty. At the same time, the ability to
1.2 – 1.3 simulate equally probable realisations of an orebody is not
1.3 – 1.4 sufficient to generate a schedule that can minimise deviations of
1.4 – 3.5 forecasts from production.

CONCLUSIONS
A generalised form of traditional mathematical modelling of
500m long-term mine scheduling for multi-element deposits has been
presented in this paper. The model was applied to a Ni laterite
deposit using a traditional, single estimated orebody model. The
application uncovers a case where the MIP formulation cannot
guarantee a feasible solution for a complex orebody. In addition,
FIG 3 - Ni grade distribution of the five simulated orebody models.
it shows that even if a mathematical model is applied
optimality is elusive due to the existence of in situ orebody
Figure 5 shows that there are significant differences between variability, which generates impractical scheduling patterns.
the scheduling patterns obtained using simulated models in direct The application also shows that traditional mathematical models
outputs of the MIP schedulers and also in the traditional schedule in mine optimisation do not account for orebody uncertainty,
illustrated in Figure 2a. These differences in the scheduling which can lead to different production schedules. For example,
patterns are caused by geological uncertainty and are the main the objective may be to have schedules that minimise risk at the
reasons in failing to meet the planned production targets in mine earlier periods of a schedule and maximise the chances of
projects. It is also clear that the schedule obtained directly from meeting production targets. Where production targets are not
the MIP model is not feasible. It requires a lot of equipment maximised, discounted cash flows will be similarly suboptimal.
movement in individual periods without considering the

Ore Tonnes
0 – 12,000
12,000 – 13,500
13,500 – 14,500
14,500 – 30,000

500m

FIG 4 - Ore tonnage distribution of the five simulated orebody models.

6 Kalgoorlie, WA, 23 - 25 April 2003 Mine Planning and Equipment Selection

 PRODUCTION SCHEDULING OPTIMISATION IN A NICKEL LATERITE DEPOSIT

1 1 2 2

3 3 4 4

5 5

Periods
1
2
3

500m

FIG 5 - The schedules obtained using traditional MIP optimisation model for five simulated orebody model inputs. The figure of each model
on the left-hand side shows the direct output of the MIP model while the one on the right-hand side shows the schedule after smoothing.

These shortfalls stem from failure to consider the required Dimitrakopoulos, R, Farrelly, C and Godoy, M C, 2002. Moving forward
smoothness of the scheduling patterns in MIP/LP-type from traditional optimisation: Grade uncertainty and risk effects in
optimisation models. The traditional MIP model schedule is open pit mine design, Trans Instn Min Metall, Sect A: Min Industry,
often too widely spread over the mineralised deposit. Too much Vol 111: A82-A89.
movement of large excavators is required, and access to blocks is Gershon, M E, 1983. Optimal mine production scheduling: Evaluation of
too restricted. In overcoming these problems during large-scale mathematical programming approaches, IJME, pp 314-329.
implementation, the mining engineer may be forced to vary Johnson, T B, 1968. Optimum open pit mine production scheduling, PhD
block scheduling and to upset ore production and blending. Thesis (University of California, Berkeley: Berkley).
Suboptimal or non-feasible MIP model outcomes may be the Johnson, T B, Dagdelen, K and Ramazan, S, 2002. Open pit mine
result. scheduling based on fundamental tree algorithm, in Proceedings 30th
International Symposium on Computer Applications in the Mineral
Industries (APCOM), (SME: Littleton).
REFERENCES Ramazan, S, 2001. Open pit mine scheduling based on fundamental tree
Akaike, A, 1999. Strategic planning of long term production schedule algorithm, PhD Thesis (Colorado School of Mines: Golden).
using 4D net-work relaxation method, PhD Thesis (Colorado School Ravenscroft, P J, 1992. Risk analysis for mine scheduling by conditional
of Mines: Golden). simulation, Trans Instn Min Metall, Sect A: Min Industry, Vol 101:
Dagdelen, K, 1985. Optimum multi-period open pit mine production A104-A108.
scheduling by lagrangian parameterization, PhD Thesis (Colorado Smith, M L and Dimitrakopoulos, R, 1999. Influence of deposit
School of Mines: Golden). modelling on mine production scheduling, International Journal of
Dimitrakopoulos, R, 1998. Conditional simulation algorithms for Surface Mining, Reclamation and Environment, Vol 13, pp 173-178.
modelling orebody uncertainty in open pit optimisation,
International Journal of Surface Mining, Reclamation and
Environment, Vol 12, pp 173-179.

Mine Planning and Equipment Selection Kalgoorlie, WA, 23 - 25 April 2003 7