SimSched Direct Block Scheduler: A new practical algorithm for the open pit

mine production scheduling problem

R. R. M. Ota1.2*, L. A. Martinez T.3
1
MiningMath., Brazil; 2Universidade Federal de Minas Gerais; 3 R&O Analytics Pty Ltd,
Australia;
*Corresponding author: [email protected]

Current practices in open pit mine planning normally challenges mine planners and mine
managers to make different decisions at different stages before achieving the generation of the best
long-term production scheduling for their projects. The objective of this thesis provides an in-
depth coverage of a novel open pit mine optimization framework, called SimSched Direct Block
Scheduler (SimSched DBS). This software allows for a more global optimization process in mine
planning, where it includes the steps of the traditional practices in its own single optimization
algorithm and provides the mine production scheduling straight from the block model.

Keywords: long-term production scheduling, heuristics, SimSched Direct Block Scheduler, single
step.

Introduction
Finding the best long term production scheduling is crucially important in surface mining projects, because, ideally,
it will deliver results grounded in criteria to maximize the net present value (NPV) and that respects, essentially,
operational constraints. Unfortunately, even after decades of extensive research in solving the Open Pit Optimization
problem, work remains to achieve this goal optimally [1]. The reason for this problem was once simply described by
Whittle [2]:

The pit outline with the highest value cannot be determined until the block values are known. The block
values are not known until the mining sequence is determined; and the mining sequence cannot be
determined unless the pit outline is available.

Due to the great computational effort to determine the exact (optimum) moment to extract each block in the entire
geologic area of interest, it was necessary to part the main problem in sub-steps, as determining these exact moments
for a sub-set of blocks, constrained by predefined pushbacks, is less challenging. However, according to Newman et
al. [6], three problematic aspects of this approach are: (1) the assumption of a fixed cutoff grade, which depends on
an arbitrary delineation between ore and waste; (2) the use of notional (and monotonically increasing) prices in
determining the nested pits; and (3) the piecemeal approach to the entire optimization problem which disregards the
temporal interaction of resource requirements.

Two main methodologies stands for this task. The first one relies on a well-accepted step-wised approach, based on
Lerchs-Grossman algorithm [3], to produce a series of nested pits, followed by scheduling algorithms that are
applied into a predefined subset of blocks, known as pushbacks. The results are analyzed and other artifices, such as
the use of stockpiles and cutoff optimization’s tools, can be used to work iteratively, evaluating the outputs and
rerunning the scheduling after each modification, in attempt to yield better production schedules. However, because
this method is strongly user dependent and each step is obtained separately, there’s no guarantee that the final result
is economically optimal, even if all tasks are executed perfectly [4].

The second approach is based on direct block scheduling (DBS), which are formulations that addresses the open-pit
production scheduling problem in a holistic and practical fashion. Johnson [5] was the first to present a model, based
on LP, able to define block destinations and extraction period under slope and mining constraints. However, his
proposal was ahead of its time, since no existent technology was able to deal with such complex methodology.
Fortunately, open-pit block sequencing is a heavily studied area [6] and computers evolved a lot in recent years as
well, turning DBS’s based techniques appealing for many authors to develop their algorithms.
Nowadays, computers can handle huge processing demands efficiently, allowing the uprising of a new era’s
algorithms, based on mixed integer programming (MIP) and heuristics, developed to deal, in a single optimization
process, increasingly complex datasets, i.e., of which accounts for more realistic constraints as conditioners for
NPV’s maximization. Despite the bigger computational challenge, new techniques and software, such as the
SimSched Direct Block Scheduler (SimSched DBS), are improving and emerging as a solid alternative to push
technology applied on mine planning.

This paper will provide a comprehensive description of the SimSched DBS software, covering its concepts, tools,
inputs, among other complexities.

SimSched Direct Block Scheduler

SimSched DBS is a commercial software (plugin of SGeMS), currently at Beta stage, from MiningMath, which is
able to assign extraction periods and destination to the blocks by means of an underlying mathematical optimization
problem. The DBS approach, implemented in SimSched DBS, provides a holistic view of the mine scheduling
problem by performing NPV maximization while constrained to capacity and geotechnical constraints and
considering the time value of money within a single optimization step, i.e., performed straight from the block model.
This means that the steps of pit optimization, pushback design and scheduling are not obtained separately, but in a
single process with mathematical criteria. In addition, the algorithm`s framework, based on MIP with heuristics, is
flexible to include any sort of additive constraints (fleet and excavation hours, metal production, average haul
distance, among others) and blending restrictions.

The SimSched DBS doesn’t rely on blocks precedence to know whether or not the block can be extracted. Instead, it
defines surfaces (mining blocks are grouped to form sets of elevations), of which accommodates a feasible solution,
to divide the production schedule into mining periods [7]. Using this approach, slope angles can be controlled, with
no error, over surfaces, rather than blocks.

Another key aspect of the SimSched DBS also relies on the incorporation of surfaces to generate scheduling plans,
which are the geometric parameters. It’s possible to set minimum values for bottom and mining width, a range for
vertical rate of advance and to import surfaces for a custom geometric restriction of the pit. Besides, SimSched DBS
allows the use of stockpiles, multiple destination routes, variable slope angle and plant recovery (block-by-block).

While this approach is theoretically better, the main issue related not only to SimSched DBS but also to all DBS
based applications stems from the computational complexity of solving these mathematical problems, which can be
increasingly large. Although the study conducted by Morales et al [8] showed that SimSched DBS scheduled 2.1
millions of blocks between 1.0 to 1.5 hours. Besides, providing mining plans that are both closer to the mine reality
(operational optimization) and with high NPV, compared to traditional methods, stands out as a challenging task.

Execution Overview
Some parameters of the mine scheduling problem are non-linear, e.g., operational parameters, and the literature
shows [1] that it’s cumbersome to model them with LP and produce results in a reasonable time. Although
SimSched DBS incorporates LP mechanisms, this process model continuous, and simpler, problems, then, along
with its own algorithms, i.e., proprietary heuristics, it interprets continuous results and transforms them, with
approximations, into integer and non-linear results. The heuristics model the problem as a whole, that is, the impacts
that a previous event can influence in further decisions are considered simultaneously, therefore, after each iteration,
the heuristics recognizes the problem and learns what the best decisions to take are.

First, SimSched DBS verifies feasibility and then optimality. The imported data and the constraints defined by the
user compose a package that will govern the survey of SimSched DBS to find and filter solutions that attend these
criteria, making this process more computationally efficient. However, achieving a solution that respects all the
parameters might not be possible in some cases, therefore, instead of not converging into a solution, it surveys other
possibilities by verifying other mathematical models, or, in case this doesn’t succeed either, by relaxing some
parameters, following a sort of hierarchy. After computing feasibility, the next step focuses on verifying the NPV.
This process will execute LP with a different model, attempting to find better NPV. In case the solution improves,
it’ll store it and run another iteration, but in case it doesn’t, it’ll verify the existence of other mathematical models.
Finally, the software halts the execution when all attempts have been tested and no improvement on the NPV is
found. Figure 1 illustrates a simplified flowchart of internal processes and decisions taken in an execution of
SimSched DBS.

Figure 1: Simplified flowchart of internal processes and decisions taken in an execution of

SimSched DBS.

Mathematical Model
The objective function present in SimSched DBS can be simplified by the following generic formulation, adapted
from Newman et al. [6]:

Where:
• b ε B: set of all blocks b.
• t ε T:set of periods within the horizon.
• d ε D: set of all destinations d.
• υbtd: discounted value associated with the final destination of a block b in period t.
• ybt: 1 if block b is extracted in period t, 0 otherwise.

The function above maximizes the NPV subjected to constraints and defines the mining sequence without being
constrained by predefined pushbacks and cutoffs, these are, instead, natural outputs of a single optimization
algorithm, executed straight from the block model.

SimSched DBS features a series of constraints that subjects the objective function. Their mathematical formulations,
among other details of the algorithm, are classified and property of MiningMath. However, it is possible to expose
generic formulations that represents the basic idea, such as the following equation for production constraints,
adapted from Newman et al. [6]:
Where:
• cb: consumption of resource associated with the extraction of block b (tons).
• C̅: minimum (maximum) resource bound in any period (tons).

The mathematical formulation found in SimSched DBS incorporates the concept of mining surfaces instead of
blocks to define the mining periods of each block. Goodwin et al. [9] and Marinho [7] are some example of authors
that presents formulations using the concept of surfaces with the motivation that this approach improves the
computational efficiency for solving the mine scheduling problem by reducing the number of variables present in
the mathematical model.

Surfaces are defined as a set of elevations, one for each coordinate of a three-dimensional orebody model, which are
used to represent the pit outline at a given mining period. More specifically, T surfaces divide each column of blocks
(fixed x and y coordinates) into T+1 group of blocks, forming T mining periods in addition to the blocks that are not
extracted. For each surface T, cell c is defined by a fixed pair of coordinates (x, y) and each cell c has an elevation
ec,t associated with period t. Variables ec,t are continuous and assume values from the origin up to the highest
elevation present in the block model. In conclusion, the surfaces govern whether a block belong to given period t by
comparing the block centroid elevation eb with the elevation ec,t of the surface, as shown in Figure 2.

Figure 2: Example of surfaces, present in SimSched DBS, defining the mining periods of the
blocks [7].

Because SimSched DBS uses the concept of surfaces, it`s possible to use the following formulation to guarantee that
no approximations are made on slope angles. This approach is also found in a stochastic scheduling model presented
by Marinho [7], who states that defining constraints over surfaces facilitates the management of geometrical
parameters as it requires fewer constraints than conventional formulations:

Figure 3: Adjacent cells elevations in x direction controlled by a maximum Hx constraint resultant

of the slope angle informed.

Beretta and Marinho [10] provide a case study to show the sensibility of slope angles approximations in open pit
mine scheduling by comparing block’s precedence method and surfaces. The study concludes that the blocks
precedence method produces bigger impacts in cases with abrupt changes in slope profiles, with variations up to
2.8% in cashflow and up to 7.8 degrees in slope angle approximations.
Constraints
SimSched DBS incorporates the package of constraints that are illustrated in Figure 4 in its optimization model, of
which also shows proximate future developments inside the dashed contoured squares. This approach is relevant to
mine planning as the solutions found are optimized and respects operational parameters, resulting in a better use of
the inputs.

Figure 4: The hierarchy of SimSched DBS constraints is presented. The image also shows
stockpiles and future implementations.

It’s possible to set minimum values for the bottom widths and mining widths. Bottom widths are the minimum
horizontal distance required to keep, in all periods, in the lowest floor of the pit, relative to adjacent slopes, to allow
mining operations. Mining widths are the minimum horizontal distance that should be excavated in every period,
which could be translated to the horizontal distance between the walls of two surfaces belonged to consecutives
periods.

The minimum widths areas are created over the surfaces by assigning the same elevation value to a set of adjacent
cells that have its centroid covered by a polygon that is resultant by the approximation of a the circle with a diameter
d, where d is equal to the parameter specified in the interface. In case the diameter d is not multiple of the grid
width, this value will be discretized to the next multiple. Figure 5 illustrates this procedure.

Figure 5: Example of a 25m minimum width defined over the surface. (A) 10x10m grid. (B)
Circle with diameter of 25m. (C) The situation is discretized to 30m. (D) A polygon is generated
 to cover all points within 30m apart. (E) An elevation ec,t is assigned to all points, creating a pit
bottom or a mining width.

The vertical rate of advance or sinking rate is defined as the vertical distance to be excavated in each period. Figure
6 shows a section view of the production scheduling of McLaughlin deposit with arrows pointing out regions where
the operational parameters can be observed.

Figure 6: Section view of McLaughlin deposit highlighting areas where the geometric parameters
can be observed.

SimSched DBS contains a functionality called “Surface Mining Limits”, which is a geometric constraint. This tool
gives the user the possibility to influence the geometry of the pit by importing surfaces as CSV files, of which
impacts on the block’s decisions over the periods. This can be done by either restricting an area to be mined until a
given period t or to force the complete exploitation in a given region in a given time frame. The limiting surfaces are
considered as strong constraints by the optimizer, meaning that they will be respected even if it implies in violating
other parameters.

The applications of the surface constraints can be vast and variable, such as to restrict areas with environmental
issues, geographic features or mining concession limits. On the other hand it can be applied, as force mining, to do
pre-stripping or to allocate in-pit crushers, as an example. In addition, the surfaces can be used to run operational
optimization, illustrated in Figure 7. The user can export the outputs of SimSched to a proper pit design software to
include other geometric features, and then import them back for a new run. As the surfaces limits are strong
constraints, the block model should be scheduled with the same surface in force and restrict mining, so that the
solution either fits inside the imported pit outline and all overlying material is mined as well. Every execution with a
different scenario configuration will yield a different solution, and users could use this in favor to work iteratively
by combining different surfaces to identify the other opportunities.

Figure 7: Simplified flowchart of how to run operational optimization with SimSched DBS. (1) An
initial solution is obtained. (2) The surfaces are exported to a pit design tool (3) Modified surfaces
are imported back to SimSched DBS in force and restrict mining. (4) New execution yields a
solution following the surfaces contour.
Stockpiles are yet not optimized, i.e., they are not an option for the optimizer to assign block`s destinations, but
rather a post-processing procedure that will search for blocks that contains marginal value that pay-offs stockpiling
costs if it’s later processed. This means that the software surveys blocks that could add value to the project if they
are later processed rather than left on waste dumps, one example of this situation is that stockpiles can contribute to
buffer productions short falls, when a processing capacity is not met only by the non-stocked material.

Blending is currently available only for the Pit Optimization mode, of which is a simplified version of the Direct
Block Scheduling mode`s algorithm that generates ultimate pits with maximized cash-flows while respecting a
series of constraints defined by the user. However, this functionality is already under development for the DBS
mode due to the relevance of this constraint for metallic open pit mine projects.

Additive constraints are in a similar situation to blending, as this functionality is available only for the Pit
Optimization mode, but are already under development for the DBS mode. They can be defined as any parameter
that the mine planners are interested to control its total “produced” amount in a given time horizon to take further
decisions, such as metal production, production by rock type, consumption of processing inputs, deforestation areas
and others. Given that, the user can define a minimum, maximum or an interval to manage the total amount
“produced” by the given parameter.

Conclusion
Since the formulation developed by Johnson [5], significant progress was achieved by the many authors that dared
to propose different formulations that include more details and provide a more realistic approach to mine planning.
However, the capacity to handle large models and provide more practical solutions, which are significant variables
to consider their applicability in real situations, arise as the main limitations of these methodologies. Given this
context, a novel tool for open pit mine optimization, SimSched Direct Block Scheduler, is pioneering in converging
its own methodologies, explored in this thesis, to handle different type of constraints in addition to the main scope
provided by all these previous studies into a concrete commercial software available for the market.

The study showed that this application is able to determine block destination and block extraction sequence
simultaneously, regarded by Onsaloo et al. [1] as an efficient way to model the problem and capable of improving
the NPV. It was seen that this approach doesn’t rely on block’s precedence, guaranteeing no violations on slope
angles, nor the assumption of a fixed cutoff grade, as the user can set customized economic values and leave that the
optimizer decides the best ore and waste delineation for each scenario. The algorithm is modeled over surfaces,
which increase efficiency by reducing the number of variables and facilitates the management of geometrical
constraints, which are the slope angles, the bottom and mining widths and the vertical rate of advance. Furthermore,
the importation of mining surfaces as constraints gives the user the possibility to influence in the final geometric
aspect of the solution, being a strong tool for operational refinements.

SimSched DBS is configured to compute the feasibility of the solution and then improve the NPV, meaning that it’ll
always deliver either one that respects all parameters or one that is partially relaxed. Due to the fact that these
considerations are modeled over the obtained solution, the discrepancies are reduced between such mathematical
projection present in SimSched DBS and what can be practically achieved, resulting in a more reliable approach.
The stochastic functionality is also present in the software and it`s an important factor to incorporate even more
fidelity to its solutions, however it wasn`t explored due not being part of the main scope of this thesis.

The software indeed is not a complete package with several tools that either aids the user`s work or permits to treat
models with certain particularities, such as calculation and script tools, optimized stockpiles, multi-pits models,
percent or sub blocked models, rotated models and others, but it is a plugin that could work from within any
software package that could already have these tools implemented. SimSched DBS is unique in its groundbreaking
way to use a single optimization algorithm, not based on Lerchs-Grossmann nor Max Flow, where the NPV
maximization is not constrained by pre-defined cut-offs, nested pits or pushbacks, these are rather natural outputs
and indicators of a complex optimization process.

Because mining projects deals with high uncertainty models and commodities prices are unstable, and with the
speed that software and hardware evolves in present era, the mining field should look after and support projects that
are willing to develop better algorithms from the scratch. Within time, the whole industry might benefit with the
investments and effort made.

References
[1] M. Osanloo, J. Gholamnejad, B. Karimi, “Long-term open pit mine production planning: a review of models
and algorithms”, International Journal of Mining Reclamation Environment. Vol. 22, no. 1, p. 3 - 35, 2008.
[2] J. Whitlle, “The Facts and Fallacies of Open Pit Optimization”, Whittle Programming Pty Ltd: North
Balwyn, Victoria, 1989.
[3] H. Lerchs, L.F. Grossmann, “Optimum Design of Open Pit Mines”, Canadian Institute of Mining Bulletin.
Vol. 58, no. 633, p. 47-54, 1965.
[4] A. Marinho, O. Guimarães, “Sequenciamento Direto de Blocos”, 8º Congresso Brasileiro de Mina a Céu
Aberto”, Belo Horizonte, 2014.
[5] T. B. Johnson, “Optimum open pit mine production scheduling”, PhD thesis, Operations Research
Department, University of California, Berkeley, 1968.
[6] A. Newman, E. Rubio, R. Caro, A. Weintraub, K. Eurek, “A review of operations research in mine planning”,
Interfaces, vol. 40, ed. 3, p. 222–245, 2010.
[7] A. Marinho, “Surface Constrained Stochastic Life-of-Mine Production Scheduling”, MSc. Thesis, McGill
University, Montreal, Qc, 2013.
[8] N. Morales, E. Jelvez, P. Nancel, A. Marinho, O. A, Guimarães, “Comparison of Conventional and Direct
Block Scheduling Methods for Open Pit Mine Production Scheduling”, Proc. APCOM 2015, Alaska. 2015.
[9] G. C. Goodwin, M. M. Seron, R. H. Middleton, M. Zhang, B. F. Hennessy, M. S. Stone, M. Menabde,
“Receding horizon control applied to optimal mine planning”, Automatica, Vol. 42 (8), pp. 1337-1342, 2005
[10] F. Beretta, A. Marinho, “Impacts of Slope Angle Approximations on Pit Optimization”, Proc. 8th Brazilian
Congress of Surface Mining, Belo Horizonte, 2014.