Scholars' Mine

Masters Theses Student Theses and Dissertations

Spring 2024

A Binary Integer Linear Programming Model for Optimizing

Underground Stope Layout
Theophilus Mensah
Missouri University of Science and Technology

Follow this and additional works at: https://scholarsmine.mst.edu/masters_theses

Part of the Mining Engineering Commons

Department:

Recommended Citation
Mensah, Theophilus, "A Binary Integer Linear Programming Model for Optimizing Underground Stope
Layout" (2024). Masters Theses. 8176.
https://scholarsmine.mst.edu/masters_theses/8176

This thesis is brought to you by Scholars' Mine, a service of the Missouri S&T Library and Learning Resources. This
work is protected by U. S. Copyright Law. Unauthorized use including reproduction for redistribution requires the
permission of the copyright holder. For more information, please contact [email protected].
A BINARY INTEGER LINEAR PROGRAMMING MODEL FOR OPTIMIZING

UNDERGROUND STOPE LAYOUT

by

THEOPHILUS MENSAH

A THESIS

Presented to the Graduate Faculty of the

MISSOURI UNIVERSITY OF SCIENCE AND TECHNOLOGY

In Partial Fulfillment of the Requirements for the Degree

MASTER OF SCIENCE

in

MINING ENGINEERING

2023

Approved by:

Kwame Awuah Offei, Advisor

Eugene Ben-Awuah
Samuel Frimpong
 © 2023

Theophilus Mensah

All Rights Reserved

 iii

ABSTRACT

Underground mine planning engineers face significant challenges when

determining what geometry provides the most profitable and safe stope for extraction.

Several techniques and optimization algorithms have been developed in recent years, but

most fail to find optimal solutions because they are heuristic or LP-based without efficient

geometric constraints. This thesis work proposes a two-dimensional binary linear

programming (BILP) model for determining the optimal combination of blocks in a stope

that maximizes the economic value of the layout of stopes for a sublevel deposit. The work

draws from Queyranne and Wolsey’s (2017 & 2018) formulations of tight constraints for

bounded up/down times in production planning problems to formulate novel and efficient

geometric constraints along with geotechnical and grade constraints for the stope layout

optimization problem. Results from the model indicate that it is possible to formulate

efficient shape constraints in LP-based approaches. The model used for the numerical

example contained 144 valuable blocks out of 774 blocks. The BILP model selected 60

valuable blocks and 13 waste blocks that met all constraints translating into a maximum

economic value of $34.4M in 1.83 hours within a gap tolerance of 0.00%. A series of

experiments show that the model is sensitive to cutoff grade, stope frame size, pillar size,

the number of stopes, and the optimization problem size. Depending on the input values

for these key parameters selected, they impact the objective function value, the solution

time and the final layout of stopes generated by the algorithm. The main limitation of the

proposed model is that pillar constraints are implemented in the Z – X or Z – Y directions

but not implemented diagonally.

 iv

ACKNOWLEDGMENTS

I would like to express my deepest gratitude to my academic advisor, Dr. Kwame

Awuah-Offei, for his invaluable guidance, support, and encouragement throughout the

course of this research. His expertise and wisdom have been instrumental in shaping the

direction and outcome of this thesis. I would also like to thank the members of my thesis

committee, Dr Samuel Frimpong and Dr Eugene Ben-Awuah, for their valuable feedback

and insights, which have greatly contributed to the improvement of this thesis.

I am also grateful to the lovely people who made my stay in Rolla feel like home.

Edie Carly Smith, Hussam Altalhi, Noah Adekunle Aluko, Eugene Gyawu, Albert

Amponsem, Joy Amponsah and Jeffery Kwarteng, your unwavering support and

encouragement throughout this journey is unmatched.

I would like to acknowledge the support and resources provided by the Mining and

Explosives Department of the Missouri University of Science and Technology, without

which this research would not have been possible.

Lastly, I thank my parents, Perpetual Oduro and John Mensah (RIP) as well as my

sister Esmiranda Mensah for their unending love, prayers, and support towards and

throughout my master’s program.

 v

TABLE OF CONTENTS

Page

ABSTRACT....................................................................................................................... iii

ACKNOWLEDGMENTS ................................................................................................. iv

LIST OF ILLUSTRATIONS ............................................................................................. ix

LIST OF TABLES ............................................................................................................ xii

NOMENCLATURE ........................................................................................................ xiv

1. INTRODUCTION ...................................................................................................... 1

1.1. BACKGROUND ................................................................................................ 1

1.2. PROBLEM STATEMENT ................................................................................. 4

1.3. OBJECTIVES AND SCOPE OF RESEARCH .................................................. 6

1.4. STRUCTURE OF THESIS ................................................................................ 7

2. LITERATURE REVIEW ........................................................................................... 8

2.1. UNDERGROUND STRATEGIC MINE PLANNING ...................................... 8

2.2. MINE PLANNING FOR UNDERGROUND STOPING METHODS .............. 9

Underground Mining Method Selection................................................. 10

Naturally Supported Underground Methods. ......................................... 11

2.2.2.1. Primary level placement. ...........................................................12

2.2.2.2. Stope layout. ..............................................................................14

2.2.2.3. Pillar support. .............................................................................15

2.2.2.4. Stope grade. ...............................................................................16

2.3. UNDERGROUND STOPE LAYOUT OPTIMIZATION ............................... 16

 vi

The Stope Layout Optimization Problem (SLOP). ................................ 19

Formulation of the SLOP with Heuristic Algorithms. ........................... 20

2.3.2.1. Simulated annealing process. .....................................................22

2.3.2.2. Genetic algorithm approach. ......................................................26

2.3.2.3. Particle swarm optimization approach. ......................................30

Formulation of The SLOP as a Linear Programming Problem. ............. 34

2.4. FORMULATION OF GEOMETRIC CONSTRAINTS IN LP

OPTIMIZATION PROBLEM ......................................................................... 38

2.5. EFFICIENT LINEAR SHAPE CONSTRAINTS ............................................ 41

2.6. SUMMARY ...................................................................................................... 44

3. BINARY INTEGER LINEAR PROGRAMMING MODELING OF

UNDERGROUND SUBLEVEL STOPE LAYOUT OPTIMIZATION ................. 46

3.1. OVERVIEW ..................................................................................................... 46

3.2. BILP MODEL FRAMEWORK ....................................................................... 46

3.3. MODEL ASSUMPTIONS ............................................................................... 48

3.4. DECISION VARIABLES, INDICES & SETS, AND PARAMETERS .......... 49

3.5. RESOURCE AND ECONOMIC MODEL ...................................................... 52

Geological Resource Model. .................................................................. 52

Economic Blockmodel. .......................................................................... 53

3.6. BILP MODEL FORMULATION .................................................................... 55

Objective Function. ................................................................................ 56

Constraints. ............................................................................................. 56

3.6.2.1. Shape constraints. ......................................................................57

3.6.2.2. Geological domain constraint. ...................................................59

 vii

3.6.2.3. Stope grade constraint. ...............................................................61

3.6.2.4. Geotechnical pillar constraints...................................................61

3.7. BILP MODEL IMPLEMENTATION .............................................................. 64

BILP Model Solution Process. ............................................................... 65

BILP Model Verification........................................................................ 65

3.7.2.1. Block economic model. .............................................................66

3.7.2.2. Stope design input......................................................................70

3.7.2.3. Results and discussions..............................................................72

3.8. SUMMARY ...................................................................................................... 77

4. EVALUATING THE MODELS SENSITIVITY TO INPUT PARAMETERS

AND OPTIMIZATION PROBLEM SIZE .............................................................. 79

4.1. OVERVIEW ..................................................................................................... 79

4.2. EFFECT OF STOPE CUTOFF GRADE ......................................................... 80

Input Data for Stope Cutoff Grade Evaluation. ...................................... 80

Results and Discussion. .......................................................................... 82

4.3. EFFECT OF STOPE DIMENSIONS (ALLOWABLE MINING

DIMENSIONS) ................................................................................................ 88

Input Data for Stope Dimension Evaluation. ......................................... 88

Results and Discussion. .......................................................................... 89

4.4. EFFECT OF STOPE PILLAR DIMENSIONS (GEOTECHNICAL

REQUIREMENT) ............................................................................................ 99

Input Data for Pillar Dimensions Evaluation. ........................................ 99

Results and Discussion. ........................................................................ 100

4.5. EFFECT OF NUMBER OF STOPES SELECTED ....................................... 106

Input Data for Number of Stopes Selected Evaluation. ....................... 106

 viii

Results and Discussion. ........................................................................ 108

4.6. EFFECT OF OPTIMIZATION PROBLEM SIZE ......................................... 114

Input Data for Optimization Problem Size Evaluation. ....................... 114

Results and Discussion. ........................................................................ 114

4.7. SUMMARY .................................................................................................... 123

5. CONCLUSIONS, RECOMMENDATIONS & FUTURE WORK ....................... 125

5.1. OVERVIEW ................................................................................................... 125

5.2. CONCLUSIONS ............................................................................................ 126

5.3. RECOMMENDATIONS FOR FUTURE WORK ......................................... 129

BIBLIOGRAPHY ............................................................................................................131

VITA ................................................................................................................................147
 ix

LIST OF ILLUSTRATIONS

Page

Figure 1.1 Typical Underground Mine [3] ......................................................................... 1

Figure 1.2 Typical Stope Layout in an Underground Mine ................................................ 2

Figure 1.3 Strategic Planning Process ................................................................................ 4

Figure 1.4 The Underground Mine Planning Problem ....................................................... 5

Figure 2.1 Underground Strategic Mine Planning Process [27] ......................................... 8

Figure 2.2 Underground Mining Methods ........................................................................ 11

Figure 2.3 Underground Sublevel Stoping Method .......................................................... 14

Figure 2.4 General Illustration of The Simulated Annealing Algorithm [86] .................. 24

Figure 2.5 Genetic Algorithm Workflow Chart [71] ........................................................ 27

Figure 2.6 A Workflow of GA Applied to Stope Layout Optimization Problem [28] ..... 28

Figure 2.7 Constraints Considered in the SLO Algorithms [57] ...................................... 40

Figure 2.8 Series of Blocks for 1D Room and Pillar [26] ................................................ 42

Figure 2.9 (a) 3×3 Rectangles and (b) 2×2 Rectangles. Arrows indicate blocks where
variables Z ivw = 1 while dark blue blocks are those that would not be in a
solution with two sets of 1D constraints but are contained in the relaxation
constraints ....................................................................................................... 44

Figure 3.1 Framework of the BILP Model ....................................................................... 47

Figure 3.2 Sample Block Model Section .......................................................................... 53

Figure 3.3 Economic Blockmodel Generation.................................................................. 54

Figure 3.4 Corner Blocks Enforcing Block Contiguity .................................................... 58

Figure 3.5 Stope Overlap Examples ................................................................................. 61

Figure 3.6 Geotechnical Rib & Sill Pillars in Stope Layout ............................................. 63
 x

Figure 3.7 The User Interface of the SSLO Application .................................................. 65

Figure 3.8 Schematic Flowchart of BILP Model Process................................................. 67

Figure 3.9 The Reblocked Resource Model ..................................................................... 68

Figure 3.10 SSLO.mlapp App Configuration ................................................................... 71

Figure 3.11 Optimal Stope Layout for Basecase Scenario ............................................... 74

Figure 3.12 Optimal Stope Layout for Basecase Scenario (Showing Stope Grades) ....... 75

Figure 3.13 The SSLO.mlapp App Showing Basecase Results........................................ 76

Figure 4.1 Experiment 1 – Optimal Layout Scenario 1 .................................................... 84

Figure 4.2 Experiment 1 – Optimal Layout Scenario 2 .................................................... 85

Figure 4.3 Experiment 1 – Optimal Layout Scenario 3 .................................................... 86

Figure 4.4 Experiment 2 Optimal Layout Scenario 1 ....................................................... 92

Figure 4.5 Experiment 2 Optimal Layout Scenario 2 ....................................................... 93

Figure 4.6 Experiment 2 Optimal Layout Scenario 3 ....................................................... 94

Figure 4.7 Experiment 2 Optimal Layout Scenario 4 ....................................................... 95

Figure 4.8 Experiment 3 Scenario 1 Optimal Layout ..................................................... 102

Figure 4.9 Experiment 3 Scenario 2 Optimal Layout ..................................................... 103

Figure 4.10 Experiment 3 Scenario 3 Optimal Layout ................................................... 104

Figure 4.11 Experiment 4 Scenario 1 Optimal Layout ................................................... 110

Figure 4.12 Experiment 4 Scenario 2 Optimal Layout ................................................... 111

Figure 4.13 Experiment 4 Scenario 3 Optimal Layout ................................................... 112

Figure 4.14 Experiment 4 Scenario 4 Optimal Layout ................................................... 113

Figure 4.15 Solution Time versus Optimization Problem Size (Number of Blocks) ..... 117
 xi

Figure 4.16 Experiment 5 Scenario 1 Optimal Layout ................................................... 119

Figure 4.17 Experiment 5 Scenario 2 Optimal Layout ................................................... 120

Figure 4.18 Experiment 5 Scenario 3 Optimal Layout ................................................... 121

 xii

LIST OF TABLES

Page

Table 2.1 Limitations of Stope Layout Optimization Algorithms .................................... 18

Table 3.1 BILP Model Sets ............................................................................................... 49

Table 3.2 BILP Model Indices .......................................................................................... 50

Table 3.3 BLIP Model Decision Variables ....................................................................... 50

Table 3.4 Technical and Economic Parameters ................................................................ 51

Table 3.5 Summary Statistics of Block Models................................................................ 69

Table 3.6 List of Economic and Technical Parameters .................................................... 69

Table 3.7 Parameters for Basecase Scenario .................................................................... 70

Table 3.8 Optimization Results for Basecase Scenario .................................................... 73

Table 4.1 BILP Input Data – Experiment 1 ...................................................................... 81

Table 4.2 Experiment 1 – Cutoff Grades .......................................................................... 81

Table 4.3 Experiment 1- Results of Changing Cutoff Grade............................................ 83

Table 4.4 BILP Input Data – Experiment 2 ...................................................................... 90

Table 4.5 Experiment 2 – Input Stope Dimensions .......................................................... 90

Table 4.6 Experiment 2- Results of Changing Stope Dimensions.................................... 91

Table 4.7 Experiment 3 – Pillar Dimensions .................................................................. 100

Table 4.8 BILP Input Data – Experiment 3 .................................................................... 100

Table 4.9 Experiment 3- Results of Changing Pillar Dimensions .................................. 101

Table 4.10 Experiment 4 – Number of Stopes ................................................................ 107

Table 4.11 BILP Input Data – Experiment 4 .................................................................. 107

Table 4.12 Experiment 4- Results of Changing Number of Stopes ................................ 109

 xiii

Table 4.13 Summary Statistics of Block Models............................................................ 115

Table 4.14 BILP Input Data – Experiment 5 .................................................................. 116

Table 4.15 Experiment 5 –Optimization Problem Size .................................................. 116

Table 4.16 Experiment 5- Results of Changing Optimization Problem Size ................. 118
 xiv

NOMENCLATURE

Symbol Description

BILP Binary integer linear programming

UG Underground
SLO Sublevel layout optimization
2D Two (2) dimensional space
LHD Load -Haul -Dump
MILP Mixed-integer linear programming
MLP Mixed-integer programming
FS Floating stope
MVN Maximum value neighborhood
MAs Metaheuristic algorithms
EAs Evolutionary algorithms
GAs Genetic algorithms
SIAs Swarm intelligence algorithms
PSO Particle swarm optimization
Z Vertical direction in blockmodel
X Horizontal direction in blockmodel

NPV Net present value

COG Cutoff grade

Sets
I Number of blocks in the Z direction in block model
J Number of blocks in the X direction in block model
K Number of stopes
W Number of pillar blocks
 xv

Indices
i = 1, 2, 3, … , I index for blocks in the Z direction in model
j = 1, 2, 3, … , J index for blocks in the X direction in model
k = 1, 2, 3, … , K index for stopes in layout
w = 1, 2, 3, … , W index for pillar blocks

Decision Variables
Binary integer variable; equals one (1) if block (i, j) is mined in
𝑥𝑖𝑗𝑘 ∈ [0, 1]
stope k; zero (0) otherwise

1
Binary integer variable; equals one (1) if block (i, j) is the topmost
𝑧𝑖𝑗𝑘 ∈ [0, 1]
block in stope k; zero (0) otherwise

2
Binary integer variable; equals one (1) if block (i, j) is the leftmost
𝑧𝑖𝑗𝑘 ∈ [0, 1]
block in stope k; zero (0) otherwise

1
Binary integer variable; equals one (1) if block (i, j) is the topmost
𝑤𝑖𝑗 ∈ [0, 1]
block of a pillar; zero (0) otherwise
Binary integer variable; equals one (1) if block (i, j) is the leftmost
𝑤𝑖𝑗2 ∈ [0, 1]
block of a pillar; zero (0) otherwise

Parameters
𝑃 Unit Price of metal
𝐶𝑠 Unit cost of refining and selling metal
𝐶𝑚𝑖𝑛 Unit cost of mining tonne of rock in block (i, j)
𝐶𝑝𝑟𝑜 Unit cost of processing a tonne of ore in block (i, j)
𝑅𝑒𝑐 Processing recovery of metal in block (i, j)
𝑔𝑖𝑗 Average grade of metal in block (i, j)
𝑇𝑖𝑗 Tonnage in block (i, j)
𝐸𝐵𝑉𝑖𝑗 Economic value of a block (i, j)
𝐺𝑜𝑓𝑓 Stope cutoff grade
𝛼1 Minimum mining height in Z-direction
 xvi

𝛼2 Minimum mining width in X-direction

𝛽1 Maximum mining height in Z-direction
𝛽2 Maximum mining width in X-direction
𝛾1 Minimum pillar length in Z-direction
𝛾2 Minimum pillar length in X-direction
 1

1. INTRODUCTION

1.1. BACKGROUND

The underground mining industry exploits deposits that are deeply seated and

economically infeasible to extract with open-pit mining techniques as illustrated below in

Figure 1.1. Complexities in geology and geotechnical characteristics of the deposits require

underground strategic mine planning engineers to select and create mine designs that are

safe, cost-effective to operate, and yield the maximum value [1]–[3].

Figure 1.1 Typical Underground Mine [3]

 2

Strategic underground mine planning involves a series of directly related activities

such as orebody modeling, stope layout design, main access network design (e.g., main

decline, ventilation raises, levels, primary drifts, and crosscuts), equipment selection, and

stope scheduling and sequencing. The first stage in value creation is the determination of

an intuitive optimal layout of extraction zones (optimal stope layout) in the deposit by

strategic mine planning engineers. This is done by using computer-aided design software.

Figure 1.2 shows the layout of stopes in an underground mine.

Figure 1.2 Typical Stope Layout in an Underground Mine

In stope-based methods (e.g., open stoping, sublevel stoping, and long hole stoping)

that require stopes for the extraction of the minerals of interest, strategic planning engineers

design several scenarios of mineable stopes based on 3D dtms/wireframe outputs from a

 3

commercial mining software (e.g., Maptek, Datamine, and Deswik) shown in Figure 1.2.

This mining software considers economic as well as technical input parameters to perform

multiple iterations that maximize the value of the mine for the investor. These input

parameters include geologic models (block models), cutoff grades, economic parameters

(e.g., metal prices, mining, and processing cost), geotechnical parameters (e.g., extraction

ratio, hydraulic ratios, and pillar widths), and geometric parameters (e.g., minimum mining

width and mining height) [4], [5].

One of the challenges of this approach is which 3D dtm/wireframe shapes and sizes

maximize value (often, at this stage, just the present value of all blocks).

Another challenge is the engines that drive this software are built off a heuristic

programming technique. Shapes are simple to model using heuristic or non-linear methods,

hence most of this commercial software includes heuristic algorithms in their optimization

packages, although these algorithms do not guarantee an optimal solution [6], [7]. However

mathematical optimization techniques such as LP-based algorithms (linear programming

and mixed integer programming) have the benefit of generating an optimal result but

modeling shapes in LP is relatively challenging due to LP's requirement that all

formulations (objective function and constraints) must be linear.

However, Queyranne [26] has shown that with the proper formulation, it is possible

to define efficient shape constraints in the LP-based algorithms that ensure contiguity and

respect rectangular shapes in the determination of the optimal stope layout. This presents

an optimization decision-making problem.

 4

Figure 1.3 Strategic Planning Process

1.2. PROBLEM STATEMENT

Mathematical programming models have been used to assist strategic planning

engineers in mine planning since the early 1960s. Compared to surface mining where there

are many models and solution algorithms [8]–[10], the underground mining environment

has fewer optimization models and solution algorithms thus the underground mine plan

optimization problem remains largely unsolved due to its complexity [10]–[12]. Studies

conducted on underground mine plan optimization suggest there are three main problems

to consider [13]–[17]:

1. Stope layout optimization

2. Access and development network optimization and

3. Stope production schedules and sequence optimization.

 5

However, the complexities surrounding the underground mine such as orebody

orientations, mining method, geotechnical characteristics, the large number of variables

and constraints to consider as well as the computational requirements, make the

underground mine plan a challenge to solve wholistically with the existing algorithms [18],

[19]. As a result, the developed algorithms only tackle one or two combinations of the

underground mine planning problem using rigorous or heuristic methodologies [20] as

shown in Figure 1.4. This thesis work will thus focus on the stope layout optimization

problem.

Figure 1.4 The Underground Mine Planning Problem

The optimal stope layout that maximizes a deposit’s value while accounting for

geotechnical requirements and grade quality constraints is the first step in the mine
 6

planning process for stoping methods. In recent years, several heuristic and linear

programming (LP) techniques and optimization algorithms have been developed in both

three-dimensional and two-dimensional space [5], [21]–[23] but most LP-based ones fail

to account for efficient shape constraints that satisfy stability and operability constraints

which leads to suboptimal solutions (i.e., even if the solution is optimal for the problem

posed, if mine engineers have to adjust that solution to implement, the implemented

solution is likely to be suboptimal). However, Queyranne’s [25], [26] work has shown that

it is possible to define efficient shape constraints that ensure continuity and respect

rectangular shapes in LP with the proper formulation.

Therefore, it is essential to develop linear programming formulations of the stope

layout problem that account for or adequately approximate such shape constraints [7], [24].

This thesis work, therefore, applies efficient shape constraints in a binary integer linear

programming formulation of the stope layout optimization problem in two-dimensional

space.

1.3. OBJECTIVES AND SCOPE OF RESEARCH

The overall objective of this thesis is to formulate the stope layout optimization

problem (SLOP) as a binary linear problem that maximizes the value of the generated

stopes subject to novel grade, geotechnical (minimum pillar sizes), and allowable mining

(minimum and maximum stope width and height) constraints in two-dimensional space.

To achieve this goal, the author:

1. Draws from Queyranne and Wolsey’s [25], [26] formulations of tight

constraints for bounded up/down times in production planning problems to

 7

formulate novel and efficient geometric constraints along with geotechnical and

grade constraints for the BILP stope layout optimization problem.

2. Illustrates the novel BILP model with a sample gold mining data set to verify

the model. The original geological model of the orebody was regularized to

generate equal-sized blocks ideal for conversion into an economic model which

serves as the primary input for the 5-experimental 15-scenario runs to verify the

BILP model as a model that applies efficient shape constraints in solving the

SLOP in two-dimensional space.

1.4. STRUCTURE OF THESIS

A comprehensive literature review on mathematical programming algorithms for

optimizing underground stope layout is covered in Section 2 of this research thesis, with

an emphasis on gaps. Section 3 presents a detailed explanation of the objective function

and constraints of the proposed BILP model for optimizing underground stope layout.

Section 4 illustrates how the BILP model was implemented on the sample data set and

discusses the findings and results deduced from the implementation. Section 5, which is

the last section of this thesis work, gives the conclusions of the study and recommendations

for future work.

 8

2. LITERATURE REVIEW

2.1. UNDERGROUND STRATEGIC MINE PLANNING

Underground strategic mine planning (SMP) is a challenging iterative process

involving a series of directly related activities including orebody modeling, stope layout

design, main access network design (e.g., main decline, ventilation raises, levels, primary

drifts, and crosscuts), equipment selection, and stope scheduling. The strategic mine

planning process for underground mining focuses on addressing three important criteria;

operational safety, profitability, and environmental stewardship. A detailed technical and

economic assessment of the deposit is required for every stage. Figure 2.1 illustrates the

iterative and cyclical nature of the underground strategic mine planning process.

Figure 2.1 Underground Strategic Mine Planning Process [27]

 9

The traditional workflow for mine planning is unable to evaluate a wide range of

options (mine plans) within a reasonable amount of time. Furthermore, with traditional

workflows, where only a few options can be considered, optimal profit is, often, not

possible. Thus, current underground strategic mine planning processes preferably rely on

the use of mathematical algorithms to handle large datasets, multiple constraints, and

numerous variables to find the optimal mine plan that maximizes value [18], [28].

Due to the complexities surrounding the underground as well as the computational

requirements, the underground strategic mine plan remains a challenge to solve

wholistically with the existing algorithms [19], [28], [29]. Consequently, the underground

strategic mine plan is subdivided into three (3) sub-problems; (1) Stope layout

optimization, (2) Access and development network optimization, and (3) Stope production

(scheduling and sequencing) optimization. The existing algorithms can tackle one or two

combinations of the underground strategic mine planning problem using rigorous or

heuristic techniques [8] as shown in Figure 1.4.

This section of the thesis focuses on reviewing the literature on heuristics, meta-

heuristic and LP-based algorithms developed in the underground mine planning space for

the stope layout optimization problem with a particular emphasis on gaps in geometric

constraints included in the formulation of these algorithms. The author reviews literature

gathered using a variety of abstracting indices including Google Scholar and OneMine.

2.2. MINE PLANNING FOR UNDERGROUND STOPING METHODS

The primary objective of mine planning is to maximize the recovery of ore while

minimizing waste rock production, ensuring the safety of workers, and minimizing the
 10

environmental impact [30]. There are multiple underground mining methods an engineer

can choose from to exploit any deposit [31]. To optimize value from the exploitation of

mineral reserves, strategic and tactical decisions regarding the most appropriate mining

method need to be made. Key considerations made in the selection of a mining method

includes host and country rock geomechanical characteristics, mineralization style,

orebody orientation, production scale, equipment, ground support systems and costs [3],

[10], [32].

Underground Mining Method Selection. The choice of an underground

mining method is an extremely important decision that affects the entire mining project.

The selection of a suitable mining method relies on sound technical and economic

evaluations of the deposit. These technical and economic analyses take into consideration

geological characteristics (dip, size, quality, and shape of the orebody), geomechanical and

geotechnical characteristics (strength of ore and host rock mass) as well as the economics

(NPV, IRR, Payback Period) of the deposit [3], [11], [33]–[36]. Generally, the definition

of a mining method permits:

• Mining equipment selection

• Stoping rate analysis

• Stope design configuration

• Mine layout configuration

Mining engineers can apply various underground mining methods to extract

mineral reserves that are located at significant depths (Figure 2.2). According to the SME

Handbook [2], some of the factors that must be considered when choosing an underground

mining method include:

 11

• Geological characteristics (extent, shape, and depth of the deposit)

• Mineralization (quality and distribution)

• Geomechanical/ Geotechnical conditions of host and country rocks.

Figure 2.2 Underground Mining Methods

The interested reader can consult mining engineering resources [2], [31], [34], [37]

to learn about the mining methods listed in Figure 2.2. In the next subsection of this work,

the thesis describes naturally supported mining methods because the stope optimization

algorithm developed in this work applies to these methods. Consequently, a background in

stoping methods allows the reader to understand the context of the work.

Naturally Supported Underground Methods. Naturally supported

underground mining methods are mining techniques that rely on the inherent stability and

strength of the surrounding rock mass to safely extract valuable minerals from beneath the

surface. These methods minimize the need for extensive artificial support systems and

instead leverage the natural characteristics of the geological formations [2], [36]. Sublevel
 12

stoping (SLS) technique for ore extraction is a prominent self-supported, selective, and

non-entry naturally supported underground mining method that is commonly used for

extracting steeply dipping, ore bodies with a thickness greater than 10 meters [2]. The

method involves splitting the orebody into horizontal tunnels known as production levels

(primary levels) with stopes separated by pillars on these levels. These primary levels can

be subdivided into sublevels for more control on production and ore selectivity. The stopes

are then mined using a mechanized system of drill & blast and haulage. The method is

particularly well-suited for mining large, low-grade deposits where high production rates

are required [37]. Figure 2.3 shows a typical layout of a sublevel stoping method.

2.2.2.1. Primary level placement. In naturally supported methods such as

sublevel stoping method, primary levels refer to the horizontal mining levels that are

established within the orebody to extract the mineralized material. These levels serve as

access points for mining activities, providing a platform for drilling, blasting, and mucking

operations [2], [37]–[39]. Here are the key considerations for determining primary levels

in the sublevel stoping method:

• Orebody geometry: Analyze the orebody's shape, size, and dip to determine the

number and spacing of primary levels. The primary levels should be positioned

at regular intervals to efficiently cover the entire orebody and ensure maximum

ore recovery.

• Vertical interval: Determine the vertical spacing between primary levels based

on the desired stope height and the mining equipment's capabilities. The vertical

interval should provide enough room for the mining operations within each

level while maintaining safe working conditions.

 13

• Access and egress: Establish primary levels at suitable locations to facilitate

efficient access to the orebody. Consider existing infrastructure, such as ramps,

declines, or shafts, to minimize the distance and cost of accessing each level.

Ensure that there are sufficient entrances and exits to accommodate personnel,

equipment, and ore transportation.

• Hanging wall and footwall stability: Consider the stability of the hanging wall

and footwall when determining the primary levels. Assess the rock mass

quality, presence of geological structures, and potential for ground instability.

Position the levels in stable rock formations to ensure the safety of workers and

equipment.

• Ventilation: Account for ventilation requirements by establishing primary

levels to allow for adequate airflow throughout the mine. Consider the direction

of airflow and position the levels to facilitate efficient ventilation and the

removal of gases, dust, and fumes generated during mining operations.

• Safety and emergency response: Ensure that escape routes, emergency exits,

and refuge chambers are appropriately positioned and easily accessible from

each level choosing the primary levels.

It's essential to note that the specific determination of primary levels may vary

based on the specific characteristics of the orebody, mining regulations, and operational

constraints. It is advisable to consult with mining professionals and engineers experienced

in sublevel stoping to develop an optimized and safe primary level layout for a particular

mining project [40], [41].

 14

Figure 2.3 Underground Sublevel Stoping Method

2.2.2.2. Stope layout. The design of the layout of a sublevel stoping method is

generally influenced by the geotechnical conditions of host and country rocks, production

scale, equipment size, and the grade of material to deliver to the processing plant [21].

These decisions determine the size, shape, and location of the stopes with respect to the

orebody. The production levels in sublevel stoping are constructed after the development

of all accesses to the stoping areas. Stopes are generated with heights ranging from 30 to

120 meters in the orebody. Stope width is generally dependent on the equipment and

orebody thickness. A raise or winze is operated into one corner of the stope from one

sublevel to the next, followed by the provision of draw points. The method can be
 15

customized to limit the number of sublevels while increasing the height of stopes between

sublevels to reduce development cost and time. A slot is then constructed for drilling and

blasting within the stope to extract the ore. Extracted ore is transported by load-haul-dump

(LHD) loaders and transferred to the underground crusher through an ore pass or to the

surface crushers using trucks, conveyor, bin, or skip. The levels above the stope crown are

protected, while neighboring stopes are separated by pillars. [37], [42].

A complete layout design of these stopes is the basis for evaluating the economic

potential of a deposit and thus the reason to ensure that an optimal layout that maximizes

value is designed [17], [30].

2.2.2.3. Pillar support. Pillar design is a crucial aspect of sublevel stoping mining

as it plays a significant role in ensuring the stability and safety of underground mining

operations. The primary purpose of pillars in sublevel stoping is to provide support to the

overlying rock mass and prevent the collapse of the stope or caving of the hanging wall

[42]. In a typical sublevel stoping operation, numerous pillars are utilized for ground

control. Rib pillars are placed as support dividers between stopes that are horizontally

contiguous. Some vertical slices are left behind as support pillars during production to help

prevent subsidence within the stope. Another important support advancement in sublevel

stopping is the use of sill and crown pillars. They are employed as a sill pillar between

vertical stopes and as the crown pillar for the transition between surface and underground

activities s [2], [33], [43], [44]. The design of pillars in sublevel stoping is influenced by a

range of factors, including rock mass quality, seismicity, ore grade, stope geometry, and

mining method [45]. The pillar size is determined based on the minimum required pillar

dimensions to ensure the stability of the overlying rock mass and prevent excessive
 16

deformation. The ratio of pillar size to stope width (hydraulic radius) is typically between

0.2 and 0.4, depending on the rock mass quality, depth of the deposit, and mining method

[45]. In general, larger pillar sizes with higher rock mass quality are necessary to maintain

stability in deep-seated deposits. In sublevel stopping mining, the optimal pillar size and

layout are commonly determined using numerical modeling and empirical techniques.

2.2.2.4. Stope grade. This Stope grade in sublevel stoping methods is a principal

factor that determines the economic viability of the mining operation. It refers to the quality

of the ore reserve material mined from a stope to meet the milling or processing

requirement., the distribution of mineralization within the deposit, the quality of the

geological and mineralization models, and The stope grade is influenced by various factors

such as the geological characteristics of the orebody (thickness, dip, strike) the distribution

of mineralization within the deposit, the quality of the geological and mineralization

models as well as dilution. Generally, the aim is to maximize the value of the project by

extracting as much ore from the deposit that exceeds the cut-off grade, while minimizing

dilution. This is achieved by selecting the best location, size, and shape of the stopes within

the orebody while ensuring geotechnical stability [2], [46], [47].

2.3. UNDERGROUND STOPE LAYOUT OPTIMIZATION

The stope layout problem involves determining the optimal arrangement of stopes

in an underground mining operation, subject to various technical, economic, and safety

constraints. In the stope layout problem, the objective is typically to maximize the net value

(NV) of the mine, which is a function of the revenue generated by the ore mined and the

costs incurred in mining and processing that ore. To do this, engineers must take into
 17

account several constraints, such as geotechnical, economic, and operational limitations

[24], [48], [49]. Traditional approaches involve manually drawing stope shapes around ore

blocks that meet the cutoff grade. This is tedious and leads to suboptimal solutions and it

is not reasonable to do all the required iterations to evaluate all possible solutions [24],

[50]–[52]. The underground stope layout problem is still difficult to solve holistically with

the current algorithms due to the complexity of underground mines, the numerous variables

to consider, and the computational requirements [19], [28], [29].

Mathematical programming models for mine planning have existed since the early

1960s with considerable development made in the surface mining space as opposed to the

underground space particularly on stope layout optimization [7], [19], [20], [53], [54].

These algorithms involve formulating an objective function that seeks to maximize value

and constraints such as ore grade, stope shape, and pillars, to find the optimal solution. The

following literature review provides an overview of the methods and techniques used for

optimizing underground stope layouts. Table 2.1 shows the main algorithms for solving

the underground stope layout problem.

 18

Table 2.1 Limitations of Stope Layout Optimization Algorithms

Dimensional
Classification Algorithm/Author(s) Optimality
Space
Octree Division (1989) [55] 3D No
Floating Stope (FS) (1995)[50] 3D No
Maximum Value Neighborhood
3D No
(MVN) (2000)[30]
Ataee-Pour (2000)[56] 3D No
Multiple Pass Floating Stope Process
3D No
(MPFS) (2001)[57]
Heuristics
Topal and Sens (2010)[58] 3D No
Sandanayake and Topal (2010)[59] 3D No
Matamoros and Kumral (2017)[60] 3D No
Nikbin et. al (2017)[61] 3D No
Sari and Kumral (2021)[44] 3D No
Clustering-Based Algorithm (2021)[63] 3D No
Dual Interchange Algorithm (2022)[46] 3D No
Dynamic Programming (1977)[64] 2D No
Downstream Geostatistics (1984)[13] 2D No
Branch and Bound (MIP)
1D Yes
(1995,1999)[55]
Probable Stope (2004)[65] 2D No
Rigorous Grieco and Dimitrakopoulos (MIP)
Not Indicated Yes
(2007)[65], [66]
Network Flow (2013)[67] 3D No
OLIPS (2007)[65] 2D Yes
GOUMA (2015)[13], [65] 2D Yes
Samanta and Suranjan (2021)[5] 3D No
 19

The Stope Layout Optimization Problem (SLOP). From literature review,

most early algorithms were purely heuristic. Heuristic algorithms, such as floating stope

(FS), maximum value neighborhood (MVN), multiple pass floating stope (MPFS), and

simulated annealing (SA), have been used to optimize stope layouts in several studies [50],

[64], [65]. Most of the early developed algorithms utilized heuristic approaches because

shapes are simple to model using heuristics or non-linear methods. Heuristic algorithms

also have the advantage of finding a good near optimality solution quickly and are based

on simple concepts or guidelines that are inspired by natural processes or engineers’

intuition. However, heuristic stope layout optimization algorithms have several

disadvantages including they result in sub-optimal solutions, they are sensitive to

parameters changes and algorithm are often complex [18], [65] [30].

Most recent algorithms are meta-heuristic to overcome the limitations of the earlier

developed heuristic algorithms. Examples include pattern search method algorithm [21],

clustering-based iterative approach [63], greedy heuristic approach [6], dual interchange

algorithm [46], simulated annealing [68], [69], and genetic algorithms [28], [70]–[72].

However, these algorithms do not guarantee optimality.

Mathematical optimization techniques, such as linear programming with its

variations such as mixed-integer linear programming (MILP) and binary integer linear

programming (BILP), provide a more efficient approach for stope layout optimization.

[20], [23], [65]. One of the main advantages of LP-based approaches is that they can handle

large-scale problems with many decision variables and constraints. They can also provide

a globally optimal solution, if the problem satisfies certain conditions, such as convexity

[73]. Several models have been developed such as mixed-integer programming (MIP) [7],
 20

[8], [53], [58], [61], [63], Network flow models [74] [73] and integer programming (IP)

[75]–[77].

However, LP-based approaches have some limitations when it comes to stope

layout optimization. One of the main limitations is that they are limited by shape constraints

[78]. In stope layout optimization, the shape of the stopes is often constrained by

geotechnical considerations, such as stability and fragmentation. These constraints can take

various forms, such as minimum width, maximum length, minimum height, and minimum

distance between adjacent stopes. Thus, it is imperative that more research is conducted to

develop effective and efficient mathematical algorithms to solve the underground stope

layout problem.

Formulation of the SLOP with Heuristic Algorithms. As stated in section

2.3.1, earlier algorithms developed to solve the SLO problem were purely heuristic with

limitations. Thus, to overcome these limitations, meta-heuristic algorithms have been

applied to most recent model developments to solve the SLO problem [79]. Metaheuristic

algorithms (MAs) are optimization algorithms that are designed to solve complex

optimization problems that are difficult to tackle with traditional optimization methods.

These algorithms are inspired by natural mechanisms and abstract concepts that aim to

efficiently explore large solution spaces to find near-optimal solutions [62], [79], [80].

MAs methods can be categorized according to various factors, including the search

strategy employed, the number of candidate solutions considered, and the extent of

hybridization or memetic techniques utilized There are several kinds of MAs, such as

Evolutionary Algorithms (EAs) and Swarm Intelligence Algorithms (SIAs).

 21

Formally, the general procedure for MAs adapted from [79] is given in the

following form:

Let X be a set of possible solutions

Find xo  X such that maximize/minimize f
(i.e. f ( xo ) = maxxX ( f ( x)) or f ( xo ) = minxX ( f ( x))
where f : X → R.

EAs the most well-known population-based, global search Mas [70], [81]. EAs are

meta-heuristic techniques that draw their inspiration from biological evolution-dependent

phenomena including reproduction, mutation, and natural selection. In EA, the search

space X is a set of chromosomes (i.e., DNA strings) regarded as candidate solutions for a

given problem. Their fitness is evaluated by objective function (f). To discover the best

solution, the fitness value of x  X should be kept as high or low as possible (i.e., the fittest

chromosome). Evolutionary algorithms (EAs) including genetic algorithms (GAs), genetic

programming (GP), evolutionary programming (EP) are examples of popular techniques

developed under the EAs.

Swarm Intelligence Algorithms (SIAs) are optimization algorithms inspired by

nature, specifically the collective behavior and interactions observed in animal colonies.

The concept of "swarm intelligence" was coined by G. Beni and J. Wang in 1989 to

describe these algorithms. Swarm Intelligence Algorithms aim to mimic the cooperative

and adaptive behavior observed in natural swarms to solve complex optimization problems.

[82].

The reason behind the widespread recognition and effectiveness of these algorithms

lies in their inherent ability to learn autonomously, their flexibility, and their capability to

adapt to changes originating from both external and internal factors [82]. In the space of
 22

stope layout optimization, several MAs have been developed to tackle the complex

constraints, large solution spaces, and the need to find a good solution within a reasonable

amount of time that characterizes the SLO problem. Examples include pattern search

method algorithm [21], clustering-based iterative approach [63], greedy heuristic approach

[6], dual interchange algorithm [46], simulated annealing [59], [60], and genetic algorithms

[28], [70]–[72]. However, these algorithms do not guarantee optimality.

In the next subsections that follows, this work will provide a good overview of the

application of these methods to the stope layout optimization problem, the general principle

behind it, their application to the SLO problem and some limitations of using these

approaches.

2.3.2.1. Simulated annealing process. Simulated annealing (SA) is a meta-

heuristic algorithm that can be used to find the global optimum solution of a non-convex

and non-linear optimization problem. The algorithm is based on the idea of simulating the

physical process of annealing in metals, where the metal is heated and then slowly cooled

to remove any defects in its structure and obtain a high-quality crystal lattice [83].

During the annealing process, a solid material, such as metal, is subjected to high

temperatures to transform it into a liquid or molten state. This elevated temperature allows

the atoms within the molten metal to move more freely. However, as the temperature is

gradually decreased, the motion of atoms becomes increasingly constrained [84].

Krikpatrik et al. [85] introduced the concept of the simulated annealing (SA)

algorithm, which can be applied to search for the global optimum of a complex function

(combinatorial problem).
 23

The algorithm starts with an initial solution and iteratively improves it by

perturbing the solution and accepting or rejecting the new solution based on a probabilistic

criterion that allows for escaping local optima [15], [69]. During each iteration of the

simulation, a fresh configuration of the system is generated by introducing a random

displacement to a randomly chosen particle from the current state. If the energy of the new

state is equal to or lower than that of the current state, the new state is immediately adopted

as the current state. However, if the energy of the new state is higher, it is still considered

for acceptance, but with a probability determined by Boltzmann's probability distribution,

Figure 2.4 shows the general SA algorithm [86].

The temperature parameter decreases over time, allowing the algorithm to escape

from local minima and converge to a near-optimal solution [69]. Although an objective

function and constraints can be used to define a problem that is addressed by simulated

annealing, in practice the constraints are incorporated into the objective function as

penalties [68], [87]. The SA algorithm can be used to optimize the stope layout by treating

the problem as a combinatorial optimization problem. In this approach, the stope layout is

represented as a binary string of 0's and 1's, where a 1 indicates that a stope is present and

a 0 indicates that it is absent. The objective function is typically the net present value (NPV)

of the mine, which is a function of the revenue generated by the extracted ore and the costs

incurred in extracting the ore and waste (that is necessary to ensure feasible stopes) as well

as the cost of processing ore.

 24

Figure 2.4 General Illustration of The Simulated Annealing Algorithm [86]

The constraints in the stope layout problem can be incorporated into the algorithm

by using penalty functions. For example, geotechnical constraints such as minimum width,

maximum length, and minimum height can be enforced by applying penalties to solutions

that violate these constraints. Similarly, the economic and operational constraints can be

incorporated into the algorithm using penalty functions [88]–[90].

 25

Simulated annealing is known for its robustness and ability to find good solutions

even when dealing with complex and non-convex optimization problems. This makes it

particularly useful in situations where other optimization methods may struggle. Also

simulated annealing is a global optimization method, meaning that it can find the global

optimum of an objective function, rather than just a local optimum. This is important in

many real-world optimization problems where finding the best possible solution is critical

[68], [91]. However, simulated annealing can be slow to converge to a satisfactory solution,

particularly for complex optimization problems such as the stope optimization problem,

which deals with many variables. This can be a disadvantage in mine planning situations

where time is a critical factor. Simulated annealing requires the user to set several

parameters, including the initial temperature and cooling rate. These parameters can be

difficult to set correctly and can have a significant impact on the performance of the

algorithm. Although simulated annealing is a global optimization method, it can still get

trapped in local minima, particularly for complex optimization problems. This can result

in suboptimal solutions. Lastly, simulated annealing can be computationally intensive,

particularly for large-scale optimization problems. This can make it impractical for certain

applications.

In summary, simulated annealing is a metaheuristic optimization method that has

found many applications in many fields including stope optimization, energy and many

more[87], [90]. However, because of its drawbacks such as the inability for SA to guarantee

an optimal solution, inability to model efficient shape constraints on the stope geometry

and its computational intensity, it has not fully addressed the problem of stope layout

optimization.
 26

2.3.2.2. Genetic algorithm approach. Another popular meta-heuristic algorithm

for stope layout optimization is the genetic algorithm (GA). Genetic algorithm (GA) is an

evolutionary inspired meta heuristic algorithm, based on the mechanism of natural

selection and biological processes of generating the fittest individual from a population

[28], [71], [92]. Applied to search for an optimal solution, GAs have the capacity to

improve solutions produced in the search space iteratively until a near optimal solution is

generated. In the 1960s and 1970s at the University of Michigan, John Holland, his

students, and colleagues pioneered and made popular the GAs [28], [81]. Since then, GAs

have grown in popularity and the diversity of applications [81], [93]–[97]

The principle behind GA mimics the process of natural selection that works on a

population consisting of competing individuals (i.e., chromosomes) where only the

strongest individuals survive. GA (Genetic Algorithms) selects a pool of parents from the

population using certain criteria, without relying on strict mathematical formulations, to

generate the next generation. As a result, GAs are considered nonlinear, discrete event, and

stochastic algorithms rather than being solely guided by mathematical rules. Crossover and

mutation operators introduce new candidates into the population. The crossover operator

creates offspring by exchanging parts of genetic information between two parents, while

the mutation operator may modify certain genes in the offspring. The elitism operator

merges the new population with the previous population and selects superior solutions from

the combined population, ensuring that performance does not deteriorate. GA assesses the

fitness of each individual using a fitness function. In the final generation, the fittest

individual is regarded as the optimal solution [81], [98]–[100].. Figure 2.5 shows the

workflow for the general principle of the GA algorithm [79], [98], [101]–[103].
 27

The GAs have been widely applied to solve complex combinatorial optimization

problems, including stope layout optimization in underground mining operations. In the

GA approach, the stope layout is represented as a chromosome in a genetic population, and

the objective function is the NPV of the mine. The geotechnical, economic, and operational

constraints can be incorporated into the algorithm using penalty functions, as in the SA

approach. Figure 2.6 shows the application of the GA to the stope layout problem [28].

Figure 2.5 Genetic Algorithm Workflow Chart [71]

 28

Figure 2.6 A Workflow of GA Applied to Stope Layout Optimization Problem [28]

28
 29

GA is a population-based metaheuristic algorithm that can effectively explore the

search space, allowing for a global search for optimal or near-optimal solutions. It has the

potential to overcome local optima and converge towards better solutions, making it

suitable for complex and multimodal optimization problems like stope layout optimization.

GA can naturally lend itself to parallel implementations, as multiple solutions can be

evaluated and evolved simultaneously. This parallelism can leverage modern computing

architectures, speeding up the optimization process and providing opportunities for

efficient utilization of computational resources.

GA can handle various constraints in stope layout optimization problems. Constraints

related to stope geometry, operational requirements, geological considerations, and others

can be incorporated into the fitness function or through customized genetic operators. GA's

ability to maintain a diverse population enhances the chance of generating feasible

solutions that adhere to the constraints.

GA also has several limitations some of which includes, tunning parameters. GA

applied to the SLOP involves several parameters that need to be carefully tuned to achieve

good performance. These parameters include the population size, crossover and mutation

rates, selection strategies, and termination criteria. Finding the optimal values for these

parameters can be challenging and often requires multiple trial-and-error iterations. Like

any metaheuristic optimization algorithm, GA is susceptible to premature convergence,

where the algorithm settles on suboptimal solutions without exploring the entire search

space. Lastly GA's optimization process may yield optimal or near-optimal solutions, but

the resulting stope layouts can be challenging to interpret and understand. The evolved

solutions may not offer clear insights into the underlying reasons for their effectiveness,
 30

making it difficult to extract actionable knowledge or recommendations for the mining

operation.

In summary, using the GA for stope optimization offers advantages such as global

search capability, solution diversity, constraint handling, flexibility, and customization.

However, it has drawbacks including parameter tuning challenges, potential for premature

convergence leading sub-optimal solution and difficulties in interpreting the results. In

SLO achieving an optimal solution is paramount to making decisions on investment as

such algorithms developed needs to ensure optimal solutions are guaranteed.

2.3.2.3. Particle swarm optimization approach. This Another meta-heuristic

technique that researchers have applied to the stope layout optimization problem is the

particle swarm optimization algorithm (PSO) [62], [80]. Particle swarm optimization

(PSO) is a metaheuristic optimization algorithm developed from swarm intelligence and

the social behavior of bird flocking or fish schooling. The first PSO was presented by

Kennedy and Eberhart to solve non-linear continuous optimization problems [104].

Particle swarm optimization is a search strategy that operates based on a population of

flying particles. These particles dynamically adjust their velocities according to their own

historical performance and the collective historical performance of the entire group, aiming

to efficiently converge towards optimal solutions in the search space[82], [105], [106].

In recent years, PSO has been widely used as an effective tool to deal with many

practical, real-life application problems such as stope layout optimization due to its ability

to handle increasingly complex problems. PSOs' popularity and success have been linked

to their capacity for self-learning, flexibility, and adaptability to both internal and external
 31

changes. as well as their ability to handle complex multi-objective problems and search for

global optimal solutions [80], [107], [82], [108].

In PSO, the technique initializes a swarm of n random particles in the search space

at random positions and velocities. The limits, the inertia factor ( w ), the cognitive and

social characteristics ( cs1 , cs2 ), and the maximum number of iterations that will be carried

out are also set during the initialization process of the algorithm. At each iteration, the

objective function value for each particle at their current position is evaluated to determine

its fitness. The particle’s best position ( x pbest ) as well as the swarm’s global best particle

position ( xgbest ) is found at this step. To move closer to the gbest and pbest particles, the

particle's current velocity and location are updated[62], [80], [82]. If any particle in the

swarm turns out to be in a position that is better than the present position of the swarm's

gbest particle, the index of the swarm's gbest particle is modified before an iteration ends.

When the stopping requirement is satisfied, that is, when the maximum number of

iterations have been finished, a good enough fitness value has been reached, or the

algorithm has been producing the same result for a period of consecutive iterations, the

iterative process is halted. The optimized function value is taken to be the fitness value of

the gbest particle at the conclusion of the process [104], [106], [108]. The formula for

updating the velocity and position [82] is, respectively, given by Equations (2.1) and (2.2):

( ) (
v n +1 = ( w  v n ) + cs1  r1  ( x pbest − x n ) + cs2  r2  ( xgbest − x n ) ) (2.1)

x n +1 = v n +1 + x n (2.2)
 32

Where:

v n +1 = velocity of the succeeding particle.

x n +1 = position of the succeeding particle.
x n = particle’s current position.
x pbest = personal best position of the particle.

xgbest = position of the global best particle of the swarm.

w = inertia factor, which controls the exploration capabilities of the algorithm
r1 and r2 are random numbers uniformly generated within the range [0,1].
cs1 and cs2 are positive parameters called the cognitive and social parameters respectively.

SI algorithms such as PSO have continuously developed over the years, leading to

a surge in research demonstrating their rapid evolution and successful implementation in

real-world optimization problems. Computational modeling of swarms using SI algorithms

has expanded beyond operations research [109] to various domains like machine learning

[110], business, and finance[110]. PSO has also been applied in engineering optimization

problems such as stope layout problem [62], scheduling and routing problems [111]. SI

algorithms are thought of as very promising optimization strategies due to the following

traits:

1. PSO is known for its the capacity to search the whole optimization problem space.

Finding global optimum or nearly optimum solutions is made possible by the

extensive range of potential solutions it investigates [112].

2. PSO’s implementation and understanding are both rather straightforward. In

comparison to other optimization methods, it contains fewer tuning parameters,

making it usable even by people with little optimization experience [111].

 33

3. PSO has shown effectiveness in solving optimization problems with a high number

of dimensions. It can handle problems with a large number of variables or decision

parameters efficiently [113].

4. Fast Convergence: PSO has the potential to converge to good solutions quickly,

especially in problems where the fitness landscape is relatively smooth and devoid

of sharp local optima [80].

However, like any other metaheuristic algorithm, the PSO has some limitations:

1. PSO does not guarantee finding the global optimum in every optimization problem.

Depending on the problem and parameter settings, PSO may converge to

suboptimal or local optima instead of the global optimum.

2. The performance of PSO is highly sensitive to its parameter settings, such as the

swarm size, inertia weight, cognitive, and social parameters. Fine-tuning these

parameters to achieve good performance can be challenging and time-consuming.

3. PSO struggles with incorporating constraints in optimization problems. Ensuring

that solutions adhere to problem-specific constraints can be challenging, requiring

additional mechanisms such as penalty functions or repair strategies.

Overall, the use of meta-heuristic algorithms in stope layout optimization can be

advantageous due to their ability to handle complex and non-linear problems, their

efficiency in searching large solution spaces, and their ability to escape local optima.

However, they have limitations in comprehensively solving the stope layout optimization

problems such as the need for tuning algorithm parameters, and the possibility that they

get stuck in a local optimum and do not find the global optimum solution.
 34

Formulation of The SLOP as a Linear Programming Problem. Linear

programming optimizes a linear objective function subject to linear equality and linear

inequality constraints. The feasible region is a convex polytope, which is the intersection

of half spaces defined by linear inequalities. The objective function is a real-valued linear

function on this polyhedron. An algorithm for linear programming identifies a point in the

polytope where the function has the smallest or largest value, if such a point exists [114]

[115], [116]. Equation (2.3) shows the general form of the LP problem with decision

variable y, and “cost coefficients” c.

Maximize cT y
 Ay  b (2.3)
subject to 
 y 0

𝒄 ∈ ℝ𝒏 , 𝒃 ∈ ℝ𝒎 are vectors and 𝑨 ∈ ℝ𝒎×𝒏 is a matrix. y ≥ 0 means that each component

of y is non-negative. Several variations of this problem are possible; for example, instead

of maximizing, we can minimize, or the constraints may be in the form of equalities, such

as Ay = b.

An instance of Equation (2.3) where all the variables and constraints are restricted

to integers is called integer linear programming (ILP) problem. A variation to this case is

when all the decision variables must be binary (i.e., 0 or 1) is called the binary integer linear

programming (BILP) problem. Mixed-integer linear programming (MILP), another

variation to the ILP is when some of the variables are restricted to integers and some

allowed to be non-integer variables [78], [117], [118].

The stope layout problem involves determining the optimal arrangement of stopes

in an underground mining operation, subject to various technical/operational (allowable

 35

mining dimensions), economic (cutoff grade), and safety constraints (pillar requirements).

The stope layout problem can, thus, be formulated as a linear programming (LP) problem,

which involves maximizing or minimizing a linear objective function subject to a set of

linear constraints [8], [14], [17], [24], [48], [55], [73].

An example of stope boundary optimization is represented as an LP formulation

as follows by Alochukwu et.al., [119]. In their model, the SLOP is formulated as a 2D

mathematical model where a binary decision variable yij is defined as yij = 1 if block (i,j)

is mined and yij = 0 if block (i,j) is not mined. The block economic value Vij is

preprocessed using a value equation similar to Equation (3.1). These two variables are then

used to define a maximization of the objective function that seeks to maximize the

economic value of the stope layout generated. The mining area is represented by an n  m

grid while the stope dimensions is a fixed    . Equations (2.4) – (2.9) summarizes their

BILP model where p is minimum stope dimension in  direction and q is minimum stope

dimension in  direction.

n− p m−q
maximise   Vij xij (2.4)
i =1 j =1

Subject to:

i+ p j +q

 x
i j
ij  1 i {1,...n − p}, j {1,...m − q} (2.5)

j +q
xij − 
j '= j +1
xij ' = 1 i {1,...n − p}, j {1,...m − q} (2.6)
 36

i+ p (2.7)
xij − x
i '=i +1
i' j = 1 i {1,...n − p}, j {1,...m − q}

i+ p j +q
xij −  
i '=i +1 j '= j +1
xi ' j ' = 1 i {1,...n − p}, j {1,...m − q} (2.8)

p =  − 1 and q =  − 1, xij  {0,1} (2.9)

This approach by Alochukwu et.al., [119] has some limitations including the use of

fixed stope dimensions, lack of flexibility to adapt to the peripheries of the deposit, and the

fact that they are not well formulated. The authors define fixed stope dimensions (    ),

which indicates the inability of this model to generate variable stope dimensions. Models

that do not give flexibility to adopt to the deposit’s peripheries will most likely generate

suboptimal shapes. The model does not integrate explicit formulations for level constraints

or geotechnical constraints, which are essential factors in effectively implementing

naturally supported stoping methods. The absence of these constraints limits the model's

ability to effectively account for the specific requirements and considerations related to

maintaining stable mining levels and addressing geotechnical challenges associated with

underground mining operations. Finally, the main problem with Alochuku et al.’s model is

that these constraints are not well formulated. Equation (2.6) – (2.8) are the constraints that

control how the model generates the stopes in the n  m grid. Equation (2.10) ensures a

block (i, j) is mined at most once in a stope. Equations (2.11) – (2.12) control the selection

of blocks in a stope (the mining constraint). The problem with these formulations for these

constraints are that they are forward looking only. Consequently if xij is mined, the blocks

ahead ((i+1) or (j+1) onwards) cannot be mined. But blocks behind it can be mined because
 37

those are not constrained, which will lead to solutions that always include the "left-hand"

side blocks only. But even then, since each block has this same constraint, only one block

can be mined in each index. Another limitation of this model is that the constraints are

formulated in the “natural” decision variables, which will lead to an exponential growth in

constraint equations leading to more computational time and resources is needed to solve

this optimization model. Time is essential in mine planning where large scale models are

used as input. Thus, such a model can negatively impact the optimization process

significantly. Finally, this model does not incorporate stope grades into the model.

Determining the material to include in a stope relies heavily on the required stope grade.

The objective is to meet the processing plant requirements by including material that

ensures the stope’s average grade meets or exceeds a specific cutoff grade. However, this

model does not incorporate stope grade constraints. As a result, some stopes formed by the

model may contain significant amounts of low-grade material since there is no control over

the threshold of grades to include in a stope.

To enhance LP models, one can introduce efficient shape constraints, tighten the

formulation of the constraints, and incorporate geotechnical and stope grade constraints.

These additions will improve the practicality and effectiveness of the model in solving the

stope layout optimization problem. This LP formulation of the stope layout problem can

be solved using standard LP solvers, such as CPLEX or Gurobi. The solution provides the

optimal selection of stopes subject to the given constraints. The LP formulation can also

be extended to incorporate additional features, such as uncertainty, multiple objectives, and

discrete variables, using appropriate modeling techniques [14], [49], [120].

 38

2.4. FORMULATION OF GEOMETRIC CONSTRAINTS IN LP OPTIMIZATION

PROBLEM

The Geometric constraints in linear programming (LP) optimization problems are

constraints that restrict the feasible region of the problem to a certain shape by imposing

limits on the decision variables. These constraints are usually applied to ensure that the

solutions to the optimization problem are physically feasible and satisfy engineering

requirements[121], [122]. The geometric constraints in LP problems can be formulated

using linear equations or non-linear equations which can be linearized [115], [123], [124].

Non-linear equations cannot be directly solved within the framework of linear

programming (LP) because LP models are based on linear relationships between decision

variables and constraints. However, non-linear equations can sometimes be linearized or

approximated to enable their inclusion in LP formulations using techniques such as the

Piecewise Linear Approximation (PLA) [115], [123]. Inequalities can also be used

depending on the specific problem and the nature of the constraints [114], [117]. Some

general methods for formulating geometric constraints in LP optimization problems are

described below.

1. Linear Inequalities: One approach to formulating geometric constraints in

LP optimization problems is to use linear inequalities[5], [24], [73]. This approach

involves specifying upper and lower bounds on the decision variables that reflect

the geometric constraints. For example, in a stope layout optimization problem, the

decision variable might be the size of a stope, and the geometric constraint might

be a minimum height and width requirement[13], [49]. This constraint could be

formulated as a linear inequality of the form:

 39

 i yi  min
(2.13)
Where:

y i = binary decision indicating if a block is included ( y i =1) or not ( y i = 0)

 i = decision variable representing stope width

 min = minimum stope width requirement

2. Nonlinear Inequalities: The geometric constraints in LP optimization

problems may be nonlinear and cannot be expressed as simple linear inequalities

[125]–[127]. However, the feasible region must still be a convex set. In such

instances, nonlinear constraints can be transformed into linear constraints using

techniques such as piecewise linearization approximation [25], [123], [128].

Linearization involves approximating a non-linear function by a linear function in

a particular region of the function's domain so the resulting linear equations can

then be used to formulate linear constraints in the LP problem [123], [124].

More complex geometric constraints, such as those involving non-rectangular stope

shapes or irregular boundaries, may require more complex formulations, such as nonlinear

equations or more complex inequalities. The specific formulation depends on the specific

problem and the nature of the geometric constraint [26], [75], [129].

Nhleko et.al [20] show that stope layout optimization problems consider several

constraints in their formulation (Figure 2.7). Their study shows none of the algorithms

developed include shape or geometry constraints in their models. LP problems do not

contain shape constraints because they are nonlinear as such the heuristics techniques tend

to be the ones that include shape constraints. Hence, most of the algorithms use heuristics

techniques [18], [65].

 40

Figure 2.7 Constraints Considered in the SLO Algorithms [57]

However, Queyranne, in his work on production sequencing and mine production

sequencing, has shown that with the proper formulation, it is possible to define efficient

shape constraints in LP-based models that ensure contiguity and respect rectangular shapes

[26]. Queyranne and Wolsey [25], [26] propose a unique approach to incorporating shape

constraints into mine planning optimization models using extended formulations. The

extended formulation approach proposed by Queyranne involves introducing additional

decision variables to identify the first block in a sequence of mined blocks. The key idea is

to reformulate the original problem in a higher-dimensional space (“natural” decision

variables), where the shape constraints can be represented by linear constraints. The

approach also provides a more compact and efficient representation of the problem. One
 41

of the main advantages of the extended formulation approach is that the resulting

optimization problem is solved using LP or MIP techniques using standard LP and MIP

solvers.

2.5. EFFICIENT LINEAR SHAPE CONSTRAINTS

Shape constraints are essential for underground mine planning because they play

an essential role in ensuring that the shapes of the stopes or mine layouts in the solutions,

meet operational and technical requirements for practical and safe extraction of the

orebody. For instance, the LHD equipment must be able to maneuver inside the stopes

while mining. Thus, not accounting for these constraints can lead to a loss of valuable ore

material, increase stope dilution, stability issues and present a suboptimal mining operation

[26]. As stated in Section 2.4 above, Queyranne and Wolsey [25], [26] presented an

approach to incorporate linear shape constraints into mine planning optimization models

using extended formulations.

The approach by Queyranne and Wolsey assumes a discrete (1D) series of blocks

as in Figure 2.8. A stope starting with block t can have length at least  t and at most  t .

Similarly, a pillar starting with block t can have length at least  t and at most  t . Their

model defines the binary decision variables:

•  t = 1, if block t is a stope block; 0 otherwise.

• Z t = 1, if block t is the leftmost of a stope. Z t = 1 if  t −1 = 0 and  t = 1.

• wt = 1, if block t is the leftmost of a pillar. wt = 1 if  t −1 = 1 and  t = 0.

 42

Figure 2.8 Series of Blocks for 1D Room and Pillar [26]

Based on these decision variables, Queyranne and Wolsey proposed the constraints

described by Equations (2.14 – 2.19) as tight MIP formulations of 1D constraints [25].

𝑍𝑡 ≥ 𝑦𝑡 − 𝑦𝑡−1 𝑡 ∈ [1, 𝑛] (2.14)

∑ 𝑍𝑢 ≤ 𝑦𝑡 𝑡 ∈ [1, 𝑛]
(2.15)
𝑢∈[0,𝑡]:
𝑢+𝛼𝑢 >𝑡

𝑡+𝛽𝑡

𝑍𝑡 ≤ ∑ (1 − 𝑦𝑢 ) 𝑡: 𝑡 ≥ 0 and 𝑡 + 𝛽𝑡 (2.16)
𝑢=𝑡+1

𝑡+𝛿𝑡

𝑤𝑡 ≤ ∑ 𝑦𝑢 𝑡: 𝑡 ≥ 0 𝑎𝑛𝑑𝑡 + 𝛿𝑡 ≤ 𝑛 (2.17)
𝑢=𝑡+1

∑ 𝑤𝑢 ≤ 1 − 𝑦𝑡 𝑡 ∈ [1, 𝑛]
(2.18)
𝑢∈[0,𝑡]:
𝑢+𝛾𝑢 >𝑡
 43

𝑦𝑡 − 𝑦𝑡−1 = 𝑧𝑡 − 𝑤𝑡 𝑡 ∈ [1, 𝑛] (2.19)

Equations (2.14) and (2.15) establishes the formation of the leftmost block. The

Equation (2.14) ensures that block t must be mined to be the leftmost point and Equation

(2.15) ensures if block t is mine but block t−1 is not, then block t must be leftmost. Equation

(2.16) ensures the contiguity control on block selection after the formation of the leftmost

block Z t . Equations (2.17) and Equation (2.18) ensure the formation of pillars between the

stopes formed. Equation (2.19) establishes the link between the variables.

This 1D approach can be extended to 2D space. To do this, we can define two sets

of 1D constraints to control blocks in each dimension. However, this approach will require

twice the number of constraints and variables. Queyranne and Wolsey [25], [26] proposed

a relaxation with two sets of 1D constraints, to reduce the complexity of the model. Each

of these 1D problems is a special case of the bounded on/off interval (pillar placement)

problem. However, this 1D relaxation leads to the formation of some blocks that are not

covered by the shape template as illustrated in Figure 2.9. In Figure 2.9 Arrows indicate

blocks where variables Z ivw = 1 while dark blue blocks are those that would not be in a

solution with two sets of 1D constraints but are contained in the relaxation constraints.

These relaxed constraints do not lead to stopes that are operational as stopes in those

solutions are “connected” rather than separated by pillars. Hence, the relaxed constraints,

while more computationally efficient, cannot be used to generate feasible stopes.

Consequently, in this thesis, the author uses the approach of two sets of 1D constraints.
 44

(a) (b)

Figure 2.9 (a) 3×3 Rectangles and (b) 2×2 Rectangles. Arrows indicate blocks where
variables Z ivw = 1 while dark blue blocks are those that would not be in a solution with
two sets of 1D constraints but are contained in the relaxation constraints

2.6. SUMMARY

This section of the thesis focused on reviewing the literature on heuristics, meta-

heuristic, and LP-based algorithms for the stope layout optimization problem. The section

also identified gaps in the literature regarding geometric constraints included in the

formulation of these algorithms.

Heuristic models include the floating stope (FS) algorithm and the maximum value

neighborhood (MVN) algorithm. The FS algorithm is limited because it produces

overlapping stopes, which does not guarantee optimality. The MVN algorithm was

developed to address this limitation, but it generates different optimal solutions based on

the chosen starting point, and hence it does not guarantee optimality.

Recent algorithms in the literature are meta-heuristic algorithms such as the pattern

search method algorithm, clustering-based iterative approach, greedy heuristic approach,

 45

dual interchange algorithm, and genetic algorithms. However, these algorithms also do not

guarantee optimality.

LP-based approaches have also been developed, and one of their main advantages

is that they can handle large-scale problems with many decision variables and constraints.

They can also provide a globally optimal solution if the problem satisfies certain conditions

such as convexity. However, LP-based approaches have some limitations, such as all the

objective function and constraints equations must be linear. Shape constraints in stope

layout optimization introduce complexity, such as minimum width, maximum length,

minimum height, and minimum distance between adjacent stopes. MILP-based approaches

have been proposed to deal with shape constraints, but they can be computationally

expensive and may not scale well to large problems. However, Queyranne has shown that

with the proper formulation, it is possible to define efficient shape constraints in LP-based

models that ensure contiguity and respect rectangular shapes.

 46

3. BINARY INTEGER LINEAR PROGRAMMING MODELING OF

UNDERGROUND SUBLEVEL STOPE LAYOUT OPTIMIZATION

3.1. OVERVIEW

This section of the work will focus on the framework of the BILP stope layout

optimization (SLO) model proposed in this thesis. The author will present the assumptions

of the framework of the algorithm as well as a detailed description of the notations,

variables, parameters, objective function, and constraints of the model. This section allows

the reader to understand the context for the BILP mathematical model applied to SLO.

3.2. BILP MODEL FRAMEWORK

The goal of this thesis work is to formulate the stope layout optimization problem

(SLOP) as a binary integer problem that maximizes the value of the mined stopes subject

to novel grade, geotechnical (minimum and maximum pillar sizes), and allowable mining

(minimum and maximum stope width and height) constraints in two-dimensional space. A

key contribution of the work is to account for geotechnical and allowable mining

constraints using efficient shape constraints.

The framework of the BILP model, as illustrated in Figure 3.1, starts by converting

a geological resource model (blockmodel) into a regularized blockmodel. This is a key

primary input for the BILP model. This regularized model contains block attributes such

as quality (ore grades), density, geotech (joints, faults), processing (recoveries) as well as

block dimensions [130], [131]. The regularized model is then converted into an economic

model using technical and economic parameters (metal price, refinery cost, mining, and
 47

processing cost) supplied by the engineer. The economic model generated at this stage is

the second key input for the BILP Model.

Optimal Solution
• Objective function value
• Binary stope layout
• Gap tolerance
• Time

Figure 3.1 Framework of the BILP Model

The next stage in the framework is to model an objective function. As stated in

Section 2.3 of this thesis, the objective in an underground stope layout optimization process

is typically to maximize the economic viability of the mine to investors [13]. Thus, the goal

of the objective function is to maximize the economic value of the deposit from mining

and processing the optimal blocks from the entire set of blocks while respecting all

constraints. The next stage in the framework is the application and consideration of

operational (allowable mining dimensions), geotechnical (pillar requirements), and

economic (cutoff grade, stope grade) constraints. This set of equations is modelled to

constrain the selected blocks to generate a feasible combination of blocks into stopes that

form the optimal layout. The last stage in the framework is the visualization and analysis

of the optimal layout of stopes.

The following subsections in this section will provide details of each stage of the

framework and provide a comprehensive description of key assumptions, primary inputs,

 48

decision variables, constraints as well as mathematical formulations developed for the

BILP model.

3.3. MODEL ASSUMPTIONS

The thesis work and the modeling effort makes several assumptions. Some of the

most critical assumptions considered are:

1. The model is limited to two-dimensional (2D) space for now to verify the

possibility of modeling effective shape constraints in LP- based algorithms used

to solve the SLO problem. However, the model is formulated in a way that

makes it possible to later extend it into 3- dimensional (3D) space.

2. The model assumes a uniform material density to simplify the formulation

although this does not cause any loss of generality (one can simply include a

tonnage factor in the formulation to account for varying block densities).

3. Binary variables were used for modeling since it establishes the decision to

include a block in the stope or not.

4. There is no cap on the number of stopes for the final design. However, the

author assumed a reasonable number to allow for multiple stopes.

5. The block model is the primary input to generate the economic block values

and it was regularized to have equal block sizes (no loss of generality because

irregular block models can always be reblocked into regular block sizes).
 49

3.4. DECISION VARIABLES, INDICES & SETS, AND PARAMETERS

This subsection provides details of the technical and economic parameters used for

the conversion of block models into economic models. It also gives a comprehensive

description of the decision variables, indices, and sets as well as the mathematical notations

used to develop the BILP model. Tables 3.1 – 3.3 contain the definitions of the notations

used for the decision variables the sets and the indices of each block in the model while

Table 3.4 contains definitions for the parameters that was used in the economic block value

calculation function.

Table 3.1 BILP Model Sets

Set Value Definition

I {1, 2, 3, … , I} Number of blocks in the Z direction in block model

J {1, 2, 3, … , J} Number of blocks in X direction in block model

K {1, 2, 3, … , K} Number of stopes

W {1, 2, 3, … , W} Number of pillars

 50

Table 3.2 BILP Model Indices

Index Value Definition

i i = 1, 2, 3, … , I index for blocks in the Z direction in model

j j = 1, 2, 3, … , J index for blocks in the X direction in model

k k = 1, 2, 3, … , K index for stopes in the layout

w w = 1, 2, 3, … , W index for the pillar blocks

Table 3.3 BLIP Model Decision Variables

Index Value Definition

1 if block (i, j) is mined in stope k;

𝑥𝑖𝑗𝑘 𝑥𝑖𝑗𝑘 ∈ [0, 1]
0 otherwise

1 if block (i, j) is the topmost block in stope k;

1 1
𝑧𝑖𝑗𝑘 𝑧𝑖𝑗𝑘 ∈ [0, 1]
0 otherwise

1 if block (i, j) is the leftmost block in stope k;

2 2
𝑧𝑖𝑗𝑘 𝑧𝑖𝑗𝑘 ∈ [0, 1]
0 otherwise

1 if block (i, j) is a topmost block of pillar w;

1 1
𝑤𝑖𝑗 𝑤𝑖𝑗 ∈ [0, 1]
0 otherwise

1 if block (i, j) is a leftmost block of pillar w;

𝑤𝑖𝑗2 𝑤𝑖𝑗2 ∈ [0, 1]
0 otherwise
 51

Table 3.4 Technical and Economic Parameters

Parameter Unit Definition

𝑃 $/𝑜𝑧 Price of metal

𝐶𝑠 $/𝑜𝑧 Cost of selling (refinery) the metal

𝐶𝑚𝑖𝑛 $/𝑡 Cost of mining a tonne of rock

𝐶𝑝𝑟𝑜 $/𝑡 Cost of processing a tonne of rock

𝑅𝑒𝑐 % Processing recovery of metal

𝑔𝑖𝑗 𝑔/𝑡 Grade of metal in a block (i, j)

𝑇𝑖𝑗 𝑡 Tonnage of block (i, j)

𝐸𝐵𝑉𝑖𝑗 $ Economic value of a block (i, j)

𝐺𝑜𝑓𝑓 𝑔/𝑡 Stope cutoff grade

𝛼1 𝑚 Minimum mining height in Z-direction

𝛼2 𝑚 Minimum mining width in X-direction

𝛽1 𝑚 Maximum mining height in Z-direction

𝛽2 𝑚 Maximum mining width in X-direction

𝛾1 𝑚 Minimum pillar length in Z-direction

𝛾2 𝑚 Minimum pillar length in X-direction

 52

3.5. RESOURCE AND ECONOMIC MODEL

This section provides a detailed description of the geological resource model and

the various mineralization domains as well as the block schema and the various attributes

that will be useful in the optimization model.

Geological Resource Model. Block modeling is an essential tool for mineral

resource estimation and mine planning in the mining industry. Block models are used to

create reliable and accurate estimates of the location, size, and quality of mineral resources

in a deposit, which is essential for assessing the economic viability of a mining project

[132]. A geological resource model (block model) is a simplified mathematical

representation of a geological deposit (ore body) and its surroundings discretized into

small, regular-shaped blocks (cells). Each block is assigned attributes such as grade,

density, and other geological and/or engineering characteristics of the mineralization

distribution within the deposit [2], [131], [133].

The block model is created using a combination of geological, geophysical, and

geochemical data collected from exploration activities, such as drilling, sampling, and

mapping. The blocks are typically defined by their x, y, and z coordinates in an XYZ grid

system, and the blocks may be of equal or of variable sizes depending on the resolution

defined by the geologist [131].

The attributes of each block, such as grade, density, and other geological

characteristics, are estimated using geostatistical and mathematical techniques, such as

kriging or inverse distance weighting, based on the available data. The block model is

typically validated using statistical and graphical methods to assess the accuracy of the
 53

model predictions and identify any areas of uncertainty [130]. Figure 3.2 illustrates a

blockmodel section.

Figure 3.2 Sample Block Model Section

Economic Blockmodel. The economic block value (EBV) is one of the most

important parameters considered in mine valuation. This parameter has considerable

impact on important decisions like the ultimate open pit (OP) limit, final UG stope layout,

the mining sequence and net present value (NPV) of a mining project. Therefore, it is
 54

necessary to calculate the EBVij at the first stage of the mine planning process, correctly.

Unrealistic economic block value estimation may cause the mining project’s managers to

make the wrong decision and may consequently subject investors to unimaginable losses

[134].

Each block within the geological block model has specific geological data, such as

grade, volume, density, and lithology. The geological data together with technical and

economic factors such as metal prices, mining cost, processing cost and mineral processing

recovery rate are then used to calculate the economic value of each block (i, j) called the

economic block value, EBVij thus converting it into an economic model. This economic

model is a key input for the BILP model [133], [134]. Figure 3.3 illustrates the conversion

of a geological blockmodel into an economic blockmodel.

Figure 3.3 Economic Blockmodel Generation

 55

Equation (3.1) is a generic mathematical formula used to estimate the EBV of each

block (i, j) in the economic block model. EBV is the undiscounted revenue from mining

and processing the block [76].

(𝐸𝐵𝑉𝑖𝑗 ) = [(𝑃 − 𝐶𝑠 ) × 𝑔𝑖𝑗 × 𝑅𝑒𝑐 − (𝐶𝑚𝑖𝑛 + 𝐶𝑝𝑟𝑜 )] × 𝑇𝑖𝑗 (3.1)

Where:

𝑃 = 𝑃𝑟𝑖𝑐𝑒 𝑜𝑓 𝑀𝑒𝑡𝑎𝑙
𝐶𝑠 = 𝑅𝑒𝑓𝑖𝑛𝑒𝑟𝑦 𝑎𝑛𝑑 𝑆𝑒𝑙𝑙𝑖𝑛𝑔 𝐶𝑜𝑠𝑡 𝑜𝑓 𝑀𝑒𝑡𝑎𝑙
𝑔𝑖𝑗 = 𝐺𝑟𝑎𝑑𝑒 𝑜𝑓 𝐵𝑙𝑜𝑐𝑘
𝑅𝑒𝑐 = 𝑃𝑟𝑜𝑐𝑒𝑠𝑠 𝑅𝑒𝑐𝑜𝑣𝑒𝑟𝑦 𝑜𝑓 𝑀𝑒𝑡𝑎𝑙
𝐶𝑚𝑖𝑛 = 𝑀𝑖𝑛𝑖𝑛𝑔 𝐶𝑜𝑠𝑡
𝐶𝑝𝑟𝑜 = 𝑃𝑟𝑜𝑐𝑒𝑠𝑠𝑖𝑛𝑔 𝐶𝑜𝑠𝑡
𝑇𝑖𝑗 = 𝑇𝑜𝑛𝑛𝑎𝑔𝑒 𝑜𝑓 𝐵𝑙𝑜𝑐𝑘

3.6. BILP MODEL FORMULATION

The BILP mathematical model applies efficient shape constraints in a binary integer

linear programming model to find the optimal combination of mining blocks into stopes

yielding the maximum value of a deposit. The work draws from Queyranne and Wolsey’s

[25], [26] formulations of tight constraints for bounded up/down times in production

planning problems to formulate novel and efficient geometric constraints along with

geotechnical and grade constraints for the stope layout optimization problem (see Section

2.5). The subsections that follow on in this section will describe the mathematical

formulations of the BILP Model. The following subsections present the objective function,
 56

the operational/technical constraints, the geotechnical constraints as well as the grade

constraints that are modelled using the decisions variables and notations.

Objective Function. The objective function of the BILP model is to

maximize the economic value (undiscounted profit) of the optimal stope layout of the

deposit. Equation (3.2) shows the objective function of this model, which is the sum of the

block values of all blocks that are selected to be included in the optimal stope layout based

on the value of the decision variable ( 𝒙𝒊𝒋𝒌 ).

𝐼 𝐽 𝐾

𝑀𝑎𝑥𝑖𝑚𝑖𝑧𝑒 ∑ ∑ ∑ 𝐸𝐵𝑉𝑖𝑗 ∗ 𝑥𝑖𝑗𝑘 (3.2)

𝑖=1 𝑗=1 𝑘=1
Where:

𝐸𝐵𝑉𝑖𝑗 = 𝐸𝑐𝑜𝑛𝑜𝑚𝑖𝑐 𝑣𝑙𝑎𝑢𝑒 𝑜𝑓 𝑏𝑙𝑜𝑐𝑘 (𝑖, 𝑗)

1 if block (i, j ) is selected to stope k

xijk = 
0 otherwise
Constraints. This section presents the constraints related to mining

requirements, geotechnical requirements, grade as well as links between the variables.

Following the example of Queyranne and Wolsey [25], [26], the author introduces new

1 2
decision variables 𝑧𝑖𝑗𝑘 and 𝑧𝑖𝑗𝑘 to represent the topmost and/or leftmost block (i, j) in a

stope k. The direct formulations of such constraints generally require exponentially many

constraints in the natural decision variables. Using this new variable enables the author to

1
model a compact constraint on the geometry forcing a more efficient stope shape. 𝑧𝑖𝑗𝑘 is

2
assigned to control the blocks along the Z-direction representing the stope height and 𝑧𝑖𝑗𝑘

is assigned to control the blocks along the X- or Y-direction depending on which section

of the deposit one uses for the optimization (for the remainder of the thesis, the work refers
 57

to the X-direction for simplicity; the reader should note the constraints are applicable to

the Y-direction in the same way if the section is in the Z-Y plane). The constraints modelled

are repeated along each coordinate direction. Other constraints are similar to previous

work and intuitive.

3.6.2.1. Shape constraints.

• Leftmost and/or Topmost Block Constraint: To form a tighter and effective

geometry the “selected" blocks must be contiguous. This work applies this

concept to define constraints that ensure that each stope k, contains a set of

contiguous blocks. To enforce this a block must be designated as the leftmost

2 1
block (𝑧𝑖𝑗𝑘 = 1) along the X-direction and topmost block (𝑧𝑖𝑗𝑘 = 1) along the

Z-direction in order to facilitate efficient formulation of the contiguity

constraints along each direction in stope k. Figure 3.4 illustrates the corner

blocks that enforce the contiguous selection of blocks (𝑥𝑖𝑗𝑘 = 1) into stope k.

Equations (3.3) and (3.4) ensure that, if block (i, j) is the leftmost or topmost

block of stope k, then the block is also mined in the stope. Equations (3.5) and

(3.6) ensures that, if block (i, j) is deemed the leftmost or topmost block in stope

k, then the preceding block (i-1, j) or (i, j-1) is not mined in that stope.

1
𝑍𝑖𝑗𝑘 ≤ 1 − 𝑥𝑖𝑗𝑘 ∀𝑖, 𝑗 , 𝑘 (3.3)

2
𝑍𝑖𝑗𝑘 ≤ 1 − 𝑥𝑖𝑗𝑘 ∀𝑖, 𝑗 , 𝑘 (3.4)

1
𝑍𝑖𝑗𝑘 ≥ 𝑥𝑖𝑗𝑘 − 𝑥(𝑖−1)𝑗𝑘 ∀𝑖, 𝑗 , 𝑘
(3.5)
 58

2 (3.6)
𝑍𝑖𝑗𝑘 ≥ 𝑥𝑖𝑗𝑘 − 𝑥𝑖(j−1)𝑘 ∀𝑖, 𝑗 , 𝑘

Figure 3.4 Corner Blocks Enforcing Block Contiguity

• Block Contiguity Constraint (Operational Requirement): Each stope must meet

a minimum and maximum mining requirement for practical extraction of the

ore. This is mine specific and varies based on geomechanical properties of the

host rock and ore as well as the scale of operation [45]. There are numerous

combinations of stope dimensions that can be used to produce a stope layout as

seen in Figure 3.4 above. This permits the mining to follow irregular mineral

deposit peripheries to minimize dilution, among other key considerations [135].

The following set of equations enables this work to model a constraint on the

stope size (height and length dimensions).  1 and  2 are the minimum

dimensions of the stopes, in the Z and X directions respectively, in units of

number of blocks. Similarly, 1 and  2 are the maximum dimensions of the

 59

stope in the Z and X directions respectively. Equations (3.7) – (3.10) ensure all

stopes meet the minimum and maximum dimensions in both directions.

𝑖
1
∑ 𝑍𝜑𝑗𝑘 ≤ 𝑥𝑖𝑗𝑘 ∀𝑖, 𝑗, 𝑘 (3.7)
𝜑=max (𝑖−𝛼1 +1,1)

𝑗
2
∑ 𝑍𝑖𝜎𝑘 ≤ 𝑥𝑖𝑗𝑘 ∀𝑖, 𝑗, 𝑘 (3.8)
𝜎=max (𝑗−𝛼2 +1,1)

min(𝐼,𝑖+𝛽1 )
1
𝛽1 − ∑ 𝑥𝛿𝑗𝑘 ≥ 𝑍𝑖𝑗𝑘 ∀𝑖, 𝑗, 𝑘 (3.9)
𝛿=min(𝐼,𝑖+1)

min(𝐽,𝑗+𝛽2 )
(3.10)
2
𝛽2 − ∑ 𝑥𝑖𝜗𝑘 ≥ 𝑍𝑖𝑗𝑘 ∀𝑖, 𝑗, 𝑘
𝜗=min(𝐽,𝑗+1)

3.6.2.2. Geological domain constraint.

• Stope Limit Constraint: Each stope k generated from the combination of block

(i, j) must be spatially unique representing a specific domain in the deposit.

Thus, the model needs a constraint to ensure only one stope is generated per

slice of blocks from the block model.

To do this, it is important to ensure the blocks mined in each contiguous

selection of blocks is isolated into its own stope so that the resulting stope

does not have multiple combinations of blocks within and avoid solutions
 60

where the cut-off grade constraints are applied across multiple “stopes”.

This requires a new set of constraints that are not based directly on

Queyranne and Wolsey’s [25], [26] work but uses the decision variables

designating the leftmost or topmost blocks to ensure efficient constraints.

This work proposes Equation (3.11), which limits the number of leftmost
2 1
𝑧𝑖𝑗𝑘 and topmost 𝑧𝑖𝑗𝑘 blocks (generally, this thesis refers to these as “corner”

blocks) to less than the sum of the maximum number of allowable blocks in

each direction ( 1 +  2 ). While this will, technically, allow multiple stopes

that exceed the minimum number of blocks constraint but for which the sum

of corner blocks is still below the sum of the maximum blocks (i.e. 1 +  2

), we find this to be rare. Additionally, Equation (3.11) results in much more

efficient constraints than an attempt to write constraints for each individual

block.

𝐾 𝐾 2
1 2
∑ 𝑍𝑖𝑗𝑘 + ∑ 𝑍𝑖𝑗𝑘 ≤ ∑ 𝛽𝑛 ∀𝑖, 𝑗 (3.11)
𝑘=1 𝑘=1 𝑛=1

• Stope Overlap Constraint: As stated in Section 3.5.1, each block representing

a portion of the mineralization in the deposit can only be mined in one stope.

This restriction prevents overlapping of the stopes that will be formed from

solutions that mine one or more blocks in multiple stopes (Figure 3.5 illustrates

the types of solution the model should avoid). Equation (3.12) ensures this

situation does not arise in the solution.

 61

∑ 𝑥𝑖𝑗𝑘 ≤ 1 ∀𝑖, 𝑗 (3.12)

𝑘=1

Figure 3.5 Stope Overlap Examples

3.6.2.3. Stope grade constraint. One aspect of stope layout optimization is the

desire to ensure that each stope meets a certain cut-off grade (𝐺𝑜𝑓𝑓 ) such that material

mined from that stope can be sent to the mill to be processed. However, this does not

mean every single block in the stope must have a grade above the cut-off grade but only

that the average grade of all blocks in a stope exceed the cut-off grade. Equation (3.13)

ensures that the average grade of the blocks in each stope k meets the set cutoff grade,

𝐺𝑜𝑓𝑓 .

𝐾
(𝑔𝑖𝑗𝑘 × 𝑥𝑖𝑗𝑘 )
∑ ≥ 𝐺𝑜𝑓𝑓 ∀𝑖, 𝑗 (3.13)
𝑥𝑖𝑗𝑘
𝑘=1

3.6.2.4. Geotechnical pillar constraints. As stated earlier in Section 2.2.2.3, the

geotechnical characteristics of the rock mass surrounding stopes play a critical role in

determining the stability and safety of underground mining operations. The operational
 62

activities (blasting and mining) from adjacent stopes, as well as the presence of geological

structures can affect the stability of the stopes generated in the layout. Therefore, it is

important to incorporate geotechnical constraints into stope layout optimization in the form

of minimum and maximum pillar dimensions to ensure a layout that maximizes ore

recovery while minimizing the risk of geotechnical failure [45], [135]. A stope layout

optimization algorithm should, therefore, include constraints to place rib pillars as support

between stopes that are horizontally contiguous and sill pillar between vertical stopes

[136].

Equations (3.14 – 3.17) are constraints that ensure the geotechnical requirement of

pillars is enforced around the stopes. Additional decision variables, wij1 and wij2 , are

introduced that controls the corner blocks of each pillar. That is, these variables become 1

if block (i, j) is the corner block of a pillar. Equations (3.14) and (3.15) controls pillar size

(minimum and minimum pillar dimensions in vertical and horizontal directions,

respectively,  1 and  2 ). Equation (3.16) and (3.17) ensures wij1 = 1 or wij2 = 1 if block (i,

j) is the leftmost or topmost block is a pillar.

𝑖 𝐾
1
∑ 𝑤𝑣𝑗 ≤ 1 − ∑ 𝑥𝑖𝑗𝑘 ∀𝑖, 𝑗 (3.14)
𝑣=max (𝑖−𝛾1 +1,1) 𝑘=1

𝑗 𝐾
2
∑ 𝑤𝑖𝜇 ≤ 1 − ∑ 𝑥𝑖𝑗𝑘 ∀𝑖, 𝑗 (3.15)
𝜇=max (𝑗−𝛾2 +1,1) 𝑘=1

𝐾 𝐾 𝐾
1 1
∑ 𝑥𝑖𝑗𝑘 − ∑ 𝑥(𝑖−1)𝑗𝑘 = ∑ 𝑧𝑖𝑗𝑘 − 𝑤𝑖𝑗 ∀𝑖, 𝑗 (3.16)
𝑘=1 𝑘=1 𝑘=1
 63

𝐾 𝐾 𝐾 (3.17)
2
∑ 𝑥𝑖𝑗𝑘 − ∑ 𝑥𝑖(j−1)𝑘 = ∑ 𝑧𝑖𝑗𝑘 − 𝑤𝑖𝑗2 ∀𝑖, 𝑗
𝑘=1 𝑘=1 𝑘=1

Figure 3.6 illustrates the types of pillars generated around the stopes in the layout using

these constraints. The pillars are respected around the stopes formed and cause the stope

shapes to also respect those pillar blocks. Note that, because the pillar constraints are

defined along the directions of the vertical and horizontal directions, pillar widths are not

maintained in the diagonal direction. This is a limitation of the formulation, and it is further

discussed in Section 4.4.

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0

0 0 0 0 0 0 0 0 0 x ijk =1 x ijk =1 x ijk =1 x ijk =1 wij 2 =1 0

0

0 x ijk =1 x ijk =1 x ijk =1 x ijk =1 wij 2 =1 0 0 x ijk =1 x ijk =1 x ijk =1 x ijk =1 x ijk =1 wij 2 =1 0
0

x ijk =1 x ijk =1 x ijk =1 x ijk =1 x ijk =1 wij 2 =1 0 0 x ijk =1 x ijk =1 x ijk =1 x ijk =1 x ijk =1 wij 2 =1 0
0
1
wij =1
x ijk =1 x ijk =1 x ijk =1 x ijk =1 x ijk =1 wij 2 =1 0 0 x ijk =1 x ijk =1 x ijk =1 x ijk =1 0 0
0 wij 2 =1

x ijk =1 x ijk =1 x ijk =1 x ijk =1 x ijk =1 wij 2 =1 0 0 wij 1 =1 wij 1 =1 wij 1 =1 wij 1 =1 0 0 0
0

wij 1 =1 wij 1 =1 wij 1 =1 wij 1 =1 wij 1 =1 0 0 0 0 0 0 0 0 0 0

0

0 0 0 0 0 x ijk =1 x ijk =1 x ijk =1 x ijk =1 x ijk =1 wij 2 =1 0 0

0 0 0

2
0 0 0 0 0 x ijk =1 x ijk =1 x ijk =1 x ijk =1 x ijk =1 wij =1 0 0
0 0 0
2
0 0 0 0 0 x ijk =1 x ijk =1 x ijk =1 x ijk =1 x ijk =1 wij =1 0 0
0 0 0

0 0 0 0 0 wij 1 =1 x ijk =1 x ijk =1 x ijk =1 x ijk =1 wij 2 =1 0 0

0 0 0

0 0 0 0 0 0 wij 1 =1 wij 1 =1 wij 1 =1 wij 1 =1 0 0 0

0 0

0 0 0 0 0 0 0 0 0 0 0 0 0
0 0

Figure 3.6 Geotechnical Rib & Sill Pillars in Stope Layout

 64

3.7. BILP MODEL IMPLEMENTATION

The model presented in this section is implemented in MATLAB. The MATLAB

code calls Gurobi to solve the optimization problem through Gurobi’s MATLAB API. A

custom application (App) called SSLO.mlapp was developed in MATLAB using the

MATLAB’s App Designer Toolkit [137].

The SSLO.mlapp has two (2) upload buttons which enables user to upload the

primary input data (BILP_Grade, BILP_EBV) into the algorithm. Users may upload pre-

processed Grade & EBV data in comma delimited (CSV) file format or a MATLAB file

format or upload the grade data and enter the economic data and allow the algorithm to

generate the EBVs of the blocks. The App interface has four (4) panels. Three (3) of the

panels (i.e., Block Counter, EBV Generator and Stope Configuration panels) accept user

input to configure the BILP algorithm, while the last panel (SSLO Results panel) is for

displaying the results of the optimization. A bulb in the lower left corner of the app

indicates the status of any ongoing activity. It turns blue when all the input data are

correctly keyed by the user and after the optimization completes. A red lamp indicates an

error during the input process as well as the optimization run. Figure 3.7 shows the

SSLO.mlapp user interface used to configure and run the BILP model for solving the SLO

problem.
 65

Figure 3.7 The User Interface of the SSLO Application

BILP Model Solution Process. As stated above the BILP Model is the

brain behind the SSLO.mlapp. Figure 3.8 illustrates the process for solving a 2D stope

layout optimization problem using the algorithm developed in this research.

BILP Model Verification. The formulated BILP model is verified with an

experimental gold deposit dataset. Maptek Vulcan software is used to reblock the original

resource block model into a regularized 15m x 15m x 30m model to avoid variable block

sizes in the resource model. Figure 3.2 and Figure 3.9 shows a section through the original

resource model and the regularized resource model respectively. The orebody for this
 66

deposit is irregularly shaped occurring between a depth of -560m to -1,100m below mean

sea level with a total mineral resource of 2,138,400 t at an average gold grade of 2.62 g/t.

Table 3.5 shows summary statistics of the resource models. This base case example used

to verify the algorithm includes 43 blocks in the X-direction and 18 blocks in the

Z-direction. We specify 20 stopes in the problem (number of specified stopes should

always be higher than what the engineer expects). Consequently, the problem results in

140,400 binary decision variables and 1,479,800 constraints.

The algorithm for the BILP model (the engine behind the SSLO.mlapp) was

implemented in MATLAB 2022b [137] environment. When the user clicks run in the

SSLO.mlapp, the app passes the user configuration to prepare the model in MATLAB (i.e.,

builds the model in MATLAB using MATLAB’s problem-based optimization workflow

commands [138]). MATLAB then calls GUROBI OPTIMIZER (Gurobi Optimizer version

10.0.0 build v10.0.0rc2) [120] through its MATLAB API to solve the optimization problem

at a gap tolerance of 0.0%. The base case example is run on a Dell Precision T5610

computer with an Intel(R) Xeon(R) Gold 5220R CPU @ 2.20GHz (8 CPUs) with a 32 GB

RAM.

3.7.2.1. Block economic model. Table 3.6 shows the mining, processing, and

economic data used to convert the reblocked model into the economic block model—the

main input for the SSLO.mlapp [52].

 67

Figure 3.8 Schematic Flowchart of BILP Model Process

 68

Figure 3.9 The Reblocked Resource Model

 69

Table 3.5 Summary Statistics of Block Models

Attribute Resource Model Reblocked Model

Metal Au Au
Number of blocks 2132 774
Blocks Au > 0 1344 144
Total mineralized material (Mt) 2,138,400.00 t 2,138,400.00 t
Maximum Au value (g/t) 37.563g/t 14.91g/t
Minimum Au value (g/t) 0.006g/t 0.145g/t
Average Au value (g/t) 2.61 g/t 2.62 g/t
Density 2.2kg/m3 2.2 kg/m3
Variance 5.76 %2 4.35 %2
Standard deviation (%) 2.40% 2.09%
Block size Varied 15m ×15m ×30m
Depth from surface 560m – 1,100m 560m – 1,100m

Table 3.6 List of Economic and Technical Parameters

Parameter Definition
Price of metal $1,500/oz

Cost of selling the metal $5/oz

Cost of mining a tonne of rock $15/t

Cost of processing a tonne of rock $10/t

Processing recovery of metal 90%

Tonnage of blockij 6,750t

Cutoff grade 1.5g/t

 70

3.7.2.2. Stope design input. Table 3.7 shows the stope design input data used for

this numerical example. These are also the main user inputs for the stope configuration

panel on the SSLO.mlapp as seen in Figure 3.10, which has the completed configuration

for the application to run.

Table 3.7 Parameters for Basecase Scenario

BILP Configuration Scenario

Minimum Mining Height 𝛼1 2

Minimum Mining Width 𝛼2 3

Maximum Mining Height 𝛽1 3

Maximum Mining Width 𝛽2 4

Minimum Pillar Length 𝛾1 2

Minimum Pillar Length 𝛾2 2

Number Of Stopes k 20

Cutoff Grade 𝐺𝑜𝑓𝑓 1.5g/t

 71

Figure 3.10 SSLO.mlapp App Configuration

 72

3.7.2.3. Results and discussions. Figure 3.11 presents the optimal stope layout

obtained by solving the BILP model for the underground deposit using Gurobi optimization

engine. Results from the test scenario summarized in Table 3.8 indicates that the model

found an optimal solution within an optimality gap tolerance of 0.0% in approximately

1.835 hrs. The model combined 60 ore blocks out of the 144 ore blocks and 13 waste blocks

into eight (8) stopes from the 20 stopes. This translated into an undiscounted value of $

34.37 million with 1,084,050 t of total mineralized material at an average grade of 3.40 g/t.

The eight (8) stopes in the final layout of the test scenario, satisfies the operational,

technical, and economic requirements. The stopes have variable heights and lengths,

indicating the power of the BILP model to adapt the stope shapes to the geological and

geotechnical characteristic of the deposit for maximum recovery. The BILP model also

selected 13 waste blocks as part of the layout to ensure it conforms with the efficient shape

constraints. All the stopes created, had overall stope grades above the cutoff (Goff = 1.5g/t).

As seen in Figure 3.12, the model generated stopes to the right side of the deposit which

has the best grades. Majority of the stopes are found in the central portion of the deposit

which has medium grades. The left side contains the low grades in the deposit hence a few

stopes were created in that zone. This indicates the power of the BILP model to pick out

blocks that maximizes the overall profit of the deposit. The scenario also verified that the

allowable geotechnical requirement (pillars) was enforced. Figure 3.13 shows the results

on the SSLO.mlapp.
 73

Table 3.8 Optimization Results for Basecase Scenario

Parameter Scenario

Number Of Stopes 8

Number Of Mined Blocks 73

Number Of Pillar Blocks 65

Objective Function Value ($) 34,373,085.19

Solution Time (Hrs.) 1.83

Gap Tolerance (%) 0.0

Minimum Stope Grade (g/t) 1.71

Maximum Stope Grade (g/t) 5.23

Average Layout Grade (g/t) 3.40

Total Layout Tonnage (tonnes) 1,084,050

 9

18

3 6

13
1
17

20

Figure 3.11 Optimal Stope Layout for Basecase Scenario

74
 75

18

3 6

13

17
20

Figure 3.12 Optimal Stope Layout for Basecase Scenario (Showing Stope Grades)

75
 76

Figure 3.13 The SSLO.mlapp App Showing Basecase Results

76
 77

3.8. SUMMARY

The model presented here proposes a method for optimizing the economic value of

a sublevel stope layout based on a binary integer linear programming formulation. A case

study of an underground gold mine has been used to successfully verify the BILP model.

A MATLAB Application, code name SSLO.mlapp, has successfully been created through

this research. The App together with the BILP model have been utilized to facilitate solving

the SLO problem. The App has been successfully validated for the basecase of the case

study. Based on results of the basecase, the following conclusions can be made:

• The results from the basecase study highlights the possibility to model shapes

in LP-based techniques for the stope layout problem. Unlike most LP-based

stope layout approaches, the proposed model accounts for efficient shape

constraints in the geometric constraints.

• The model finds the optimal stope layout that maximizes the undiscounted

profit for the deposit within a gap tolerance of 0.00%.

• The model allows the generation of variable stope length and height as well as

incorporating geotechnical pillar requirements between the selected stopes.

Thus, the model permits mining operation to follow irregular mineral deposit

peripheries to minimize dilution.

• The BILP model has some limitations as seen from the optimal layout in Figure

3.11. Pillars are respected around the stopes however pillar widths are not

maintained in the diagonal direction because the pillar constraints are defined

along the vertical and horizontal directions.

 78

• Lastly the use of binary variables makes the problem difficult to solve because

it creates a combinatorial explosion of possible solutions as the number of

variables increases, leading to long solution times.

 79

4. EVALUATING THE MODELS SENSITIVITY TO INPUT PARAMETERS

AND OPTIMIZATION PROBLEM SIZE

4.1. OVERVIEW

This section of the thesis describes computational experiments carried out to assess

the performance and sensitivity of the proposed BILP model’s solution to key input

parameters. The key parameters of the model such as the stope dimensions, stope cutoff

grade, the number of predefined stopes, the pillar dimensions as well as the size of the

optimization problem were varied to evaluate their impact on the solution time, the

objective function value, the optimality gap, and the final stope layout.

This study looks at the instance of changing the scale of the optimization problem

since an increase in problem size leads to an explosion in variables and, from a practical

standpoint, strategic mine engineers are unlikely to adopt an algorithm that takes more than

a few minutes to converge to a solution. The work also examined the impact of different

configurations of the stope and pillar dimensions because one of the strengths of this model

is that it allows engineers to generate any rectangular mining dimension that mimics the

deposit’s peripheries.

Stope cutoff grade is a key factor that determines the quality and quantity of

material that can be sent from the UG to the processing plant. Therefore, evaluating and

understanding the impact of this parameter will help engineers to select an optimal cutoff

grade that meets the processing plant requirement and overall project profitability. The

number of stopes is set by the engineer as an input to the algorithm. This parameter

determines the stopes that are formed in the final layout as such the author studies the effect
 80

of changing the number of stopes on solution time and overall profitability of the layout

formed.

The subsections below present the computational experiments for the different

scenarios run to evaluate the performance and sensitivity of the model. For each

experiment, the author generated economic block values and block grades in the same way

as for the base case in Section 3.7.2. The author used the same MATLAB code used in the

base case study in Section 3.7.1 to solve all the different experimental runs. The full input

data used for these experiments are available online in this GitHub repository

(https://github.com/TheoMensah/BILP_SSLO).

4.2. EFFECT OF STOPE CUTOFF GRADE

Due to the highly selective nature of underground mining, all material hauled from

the stopes generated must be at or above the predetermined cutoff grade1. This is to ensure

maximum recovery of the deposit.

Input Data for Stope Cutoff Grade Evaluation. A low cutoff grade results

in more ore tonnage, and overall higher metal output, but at the expense of additional

capital cost while a high cutoff grade, denotes a short life of mine and lower overall metal

output, which in most instance cannot be sufficient to justify the capital cost of establishing

a mine [139]. Thus, the cutoff grade selected as an input for this model must but optimal.

Techniques like MIP [140], [141] can be used to find the optimal cutoff grade for use in

the BILP model.

1
Cutoff grade is the grade value below which blocks in the deposit are uneconomical to mine.
 81

Table 4.1 and Table 4.2 show the input data used to configure the BILP model for

changing the cutoff grade. Three (3) scenarios were run for this experiment to investigate

the effect of a lower and higher cutoff grade value, than the base case value. For this

analysis, the stope configurations for each scenario for mining and geotechnical

dimensions are selected to be the same as shown in Table 4.1 while Table 4.2 summarizes

different cutoff grade values used for this experiment.

Table 4.1 BILP Input Data – Experiment 1

BILP Configuration Scenario

Minimum Mining Height 𝛼1 3

Minimum Mining Width 𝛼2 3

Maximum Mining Height 𝛽1 4

Maximum Mining Width 𝛽2 4

Minimum Pillar Length 𝛾1 2

Minimum Pillar Length 𝛾2 2

Number Of Stopes k 20

Table 4.2 Experiment 1 – Cutoff Grades

BILP Configuration Scenario 1 Scenario 2 Scenario 3

Cutoff Grade 𝐺𝑜𝑓𝑓 1.5g/t 2.5g/t 3.5g/t
 82

Results and Discussion. Figure 4.1 – Figure 4.3 show the results of this

experiment while Table 4.3 summarizes the results. The results show that the objective

function value decreases as the cutoff grade increases from $33M to $28M. This results

also show a higher cutoff grade causes the algorithm to converge faster to a solution.

Scenario 3 recorded the fastest solution time of 0.11 hrs. All scenarios achieved an optimal

solution2.

From the results the lower the cutoff grade selected, the lower the average overall

layout grade achieved however this leads to more blocks being selected and more stopes

being formed in the optimal layout and consequently maximizing the economic value. The

results also suggest that increasing cutoff grade does not improve the overall profitability

of the designed layout although it leads to higher grade stopes.

The solutions show that five (5) stopes were formed in the optimal layout for

Scenario 1 while four (4) stopes were formed in the optimal layout for Scenarios 2 and 3,

respectively (Figure 4.1 – Figure 4.3). Each stope generated in the optimal layout had an

average grade above the cutoff grade. The various geological domains of the blockmodel

used have been described earlier in Section 3.7.2. Based on this understanding, Figure 4.1

– Figure 4.3 show the algorithm produced stopes aimed at the blocks in the central and

upper right zones in each Scenario indicating the BILP models’ power to target, select and

combine blocks that maximize the value of a deposit.

2
Optimal solution means a solution with an optimality gap of 0.00%
 83

Table 4.3 Experiment 1- Results of Changing Cutoff Grade

Optimization Results
Parameter Units
Scenario 1 Scenario 2 Scenario 3
Objective Function Value ($) 33,727,574.0 30,959,129.3 28,468,130.7

Solution Time (hrs) 1.03 0.67 0.11

Gap Tolerance (%) 0.00 0.00 0.00

Number of Stopes (#) 5 4 4

Ore Blocks Mined (#) 70 57 46

Waste Blocks Mined (#) 5 3 2

Number of mined blocks (#) 75 60 48

Number of pillar blocks (#) 65 60 55

Minimum Stope Grade (g/t) 2.03 2.99 3.53

Maximum Stope Grade (g/t) 4.03 4.03 4.03

Average Layout Grade (g/t) 3.08 3.41 3.79

Total Layout Tonnage (tonnes) 1,113,750 891,000 712,800

83
 84

Figure 4.1 Experiment 1 – Optimal Layout Scenario 1

84
 85

Figure 4.2 Experiment 1 – Optimal Layout Scenario 2

85
 86

Figure 4.3 Experiment 1 – Optimal Layout Scenario 3

86
 87

From Figure 4.1, the optimal layout for Scenario 1 (which specified a stope cutoff

grade of 1.5 g/t) had three (3) stopes generated within the low-grade domain of

2g/t  g ij  3g/t , one (1) stope within the medium-grade domain of 3g/t  g ij  4g/t and

one (1) stope in the high-grade domain 4g/t  g ij  10g/t .

All stopes formed had at least 14 blocks selected and a maximum of 16 blocks

satisfying the mining requirements configured for this scenario. Since the cutoff grade is

low, the algorithm takes a longer time to optimize the numerous possible block

combinations thus leading to a longer solution time to find the optimal solution.

The optimal layout for Scenario 2 (which specified a stope cutoff grade of 2.5 g/t),

shown in Figure 4.2, had one (1) stope generated within the low-grade domain of

2g/t  g ij  3g/t , two (2) stopes within the medium-grade domain of 3g/t  g ij  4g/t and

one (1) stope in the high-grade domain 4g/t  g ij  10g/t . This layout was generated

because of the higher stope cutoff grade. All stopes formed had at least 15 blocks selected

and a maximum of 16 blocks satisfying the mining requirements configured for this

scenario.

Figure 4.3 shows the optimal layout for the highest stope cutoff grade ( Goff = 3.5g/t

) of the three scenarios. The optimal layout had three (3) stopes generated within the

medium-grade domain of 3g/t  g ij  4g/t and one (1) stope in the high-grade domain

4g/t  g ij  10g/t . No stopes were generated in the low-grade domain. By further elevating

the cutoff grade in this scenario, the algorithm went in to select the few blocks with very

high grades in the blockmodel that when combined will achieve the elevated cutoff grade
 88

( Goff = 3.5g/t ) as seen in Figure 4.3. Elevating the cutoff grade also resulted in the fastest

solution time since few blocks could meet this requirement. Thus, two (2) stopes formed

had 9 blocks selected and the other stopes had a maximum of 16 blocks satisfying the

mining requirements configured for this scenario.

4.3. EFFECT OF STOPE DIMENSIONS (ALLOWABLE MINING DIMENSIONS)

Each designed stope must meet an allowable minimum and maximum mining

dimension based on the geomechanical properties of the deposit and equipment sizes for

practical extraction of material from the stopes. The experiment was designed to investigate

the impact of frame sizes while keeping pillar sizes constant.

Input Data for Stope Dimension Evaluation. To ensure that stopes formed

follow the deposit peripheries and minimize dilution while ensuring stability and

operability, mine engineers using stope optimization algorithms should be able to control

the generated stope shapes. This is one strength of the BILP model proposed in this work

as it introduces efficient shape constraints as described in Section 3.6.2.1 that allows for

any rectangular dimension. The shape constraints in this model, allows the engineer to

control the minimum and maximum blocks for the stope in the Z-X plane or Z-Y plane.

This section investigates the sensitivity of the model to changes in stope

dimensions. The author designed the experiment to investigate the impact of frame sizes

while keeping pillar sizes constant. The reader should note that, practically, larger frame3

sizes are likely to go with larger pillar sizes (more information in Section 2.2.2.3).

3
Frame is the rectangular dimension of the stope generated in the layout.
 89

By maintaining uniform pillar dimensions, the focus was solely on understanding how

different frame configurations influence the outcomes. This approach allowed for a more

precise evaluation of the frame size's individual contribution to the results, without the

confounding effect of varying pillar sizes.

The analysis run four (4) scenarios for this experiment. The first scenario runs a

smaller stope frame ( 2  3 | 2  3 ) for  i and  i , respectively. In Scenario 2 the author

investigates a square stope frame ( 3  3 | 3  3 ) for  i and  i , respectively. The third

scenario runs an adjustment to the maximum allowable mining (  i ) dimension of the stope

frame in Scenario 2 increasing it to large rectangular ( 3  3 | 4  4 ) frame to investigate the

change in increasing the  i for the frame and the last Scenario 4, evaluates the effect of

configuring an even larger stope frame ( 3  5 | 4  6 ). Table 4.4 shows the common input

configuration for the BILP model while Table 4.5 shows different stope frame dimensions

used for this evaluation.

Results and Discussion. Table 4.6 summarizes the results of this experiment

and Figure 4.4 – Figure 4.7 shows the optimal layouts for these scenarios. From the results,

changing the allowable mining dimensions impacts the optimal solution significantly. The

stope frame configuration the engineer selects affects the objective function value, the

solution time, the number of stopes formed, and general stope layout.

From the results, the objective function value improves with an increase in the stope

dimensions although the overall layout grade of material in the stopes decreases with this

increase in stope dimensions. The reason for the higher objective function values observed

with increasing the frame size is the ability to select a greater number of oreblocks within
 90

the optimal layout. This selection of additional oreblocks maximizes the economic value

of the mining operation.

Table 4.4 BILP Input Data – Experiment 2

BILP Configuration Scenario

Minimum Pillar Length 𝛾1 2

Minimum Pillar Length 𝛾2 2

Number Of Stopes k 20

Cutoff Grade 𝐺𝑜𝑓𝑓 2g/t

Table 4.5 Experiment 2 – Input Stope Dimensions

BILP Configuration Scenario 1 Scenario 2 Scenario 3 Scenario 4

Minimum Mining Height 𝛼1 2 3 3 3

Minimum Mining Width 𝛼2 3 3 3 5

Maximum Mining Height 𝛽1 2 3 4 4

Maximum Mining Width 𝛽2 3 3 4 6

 87

Table 4.6 Experiment 2- Results of Changing Stope Dimensions

Optimization Results
Parameter Units
Scenario 1 Scenario 2 Scenario 3 Scenario 4
Objective Function Value ($) 25,581,135.8 27,455,739.1 33,727,578.4 34,740,592.9

Solution Time (hrs) 4.50 3.37 1.78 1.03

Gap Tolerance (%) 0.00 0.00 0.00 0.00

Number Of Stopes (#) 7 5 5 4

Ore Blocks Mined (#) 38 43 69 69

Waste Blocks Mined (#) 4 2 6 12

Number Of Mined Blocks (#) 42 45 75 81

Number Of Pillar Blocks (#) 55 45 35 25

Min Stope Grade (g/t) 2.01 2.49 2.03 2.63

Max Stope Grade (g/t) 7.21 5.52 4.03 3.31

Average Layout Grade (g/t) 3.85 3.86 3.08 2.98

Total Layout Tonnage (tonnes) 623,700 668,250 1,113,750 1,202,850

91
 92

Figure 4.4 Experiment 2 Optimal Layout Scenario 1

92
 93

Figure 4.5 Experiment 2 Optimal Layout Scenario 2

93
 94

Figure 4.6 Experiment 2 Optimal Layout Scenario 3

94
 95

Figure 4.7 Experiment 2 Optimal Layout Scenario 4

95
 96

It is essential to acknowledge that, in practice, larger frame sizes are often

accompanied by larger pillar sizes to ensure stability in the layout and during stoping

operations. Consequently, the results of this experiment may not be “optimal” in a practical

sense where pillar sizes vary with stope dimensions. However, it is worth noting that, in

this experiment, pillar sizes were deliberately kept constant to isolate and assess the sole

impact of frame sizes. This approach allowed for a detailed examination of how varying

frame sizes alone influence the outcomes, independent of changes in pillar dimensions.

Table 4.6 shows that if a smaller frame configuration is selected, the algorithm takes

a longer time to converge to an optimal solution. This extended time is primarily attributed

to the increased selectivity achieved with a smaller frame. Therefore, with smaller frame

sizes, the options to evaluate increase and the number of constraints also increase, leading

to longer computational times.

In each scenario, at least four stopes were created, but the number fell as the stope

size increased. All the stopes generated targeted blocks in the deposit's upper right and

central zones. The optimal layouts shown in Figure 4.4 – Figure 4.7 produced variable

stope heights and lengths that matched the input data in Table 4.5. To respect the shape

constraints a few waste blocks are selected as part of the optimal layout. These blocks must

be carefully selected as they introduce internal dilution to the generated stopes and reduce

the overall value. Thus, one strength of the BILP model is the ability to optimally select

waste blocks as part of the optimal stope generated but ensure that the stope cutoff grade

and the shape constraints are respected by the solution. The number of waste blocks

selected in the final layout also increases with an increase in the stope frame dimensions.

The mining dimensions are site specific and varies from mine to mine thus a good
 97

geological and geomechanical database of the deposit is recommended when determining

the stope dimension to select for optimization.

In Scenario 1, seven (7) stopes were generated in the optimal layout, each

conforming to the frame ( 2  3 | 2  3 ) configured for the scenario. This solution included

42 blocks; 38 ore blocks from a possible 144 oreblocks and 4 waste blocks4 to complete

the layout. This translated into an economic value of $26 M with total mineralized material

of 623 Kt at 3.85 g/t. The small size of this stope frame means more stopes will be formed;

however, due to the variability of grades in the deposit and the requirement of pillars, not

all blocks can be included in the optimal layout of stopes. Thus, to ensure the economic

value is maximized, the algorithm targets blocks with medium to high grades (Figure 4.4).

The pillar constraints are only respected in the X and Z-directions (not along the

diagonal directions). Some of these stopes in our solution will be unsafe to mine because

they are next to each other diagonally. Stopes 3 – 17 – 6 and stopes 9 – 6, in Figure 4.4,

illustrate this issue. Section 3.8 discusses this limitation of the proposed BILP model. One

possible solution to this problem is to post-process the solution with heuristics to avoid

these situations. Another possible solution will be to model sill pillars into the layout. This

will separate the stopes into levels and can potentially eliminate the direct diagonal

interaction of some of the stopes generated in the layout.

Figure 4.5 illustrates the optimal layout of stopes for Scenario 2 which has

( 3  3 | 3  3 ) stope frames. From the layout in Figure 4.5, given the stope cutoff grade

4
Waste blocks are included as internal dilutions and to ensure stopes respect the shape constraints.
 98

( Goff = 2g/t ) and the square frame, five (5) stopes were formed each with 9 blocks

(min/max of 3 blocks in the Z-direction and 3 blocks in the X-direction) conforming to the

square frame configuration.

Figure 4.6 illustrates the optimal layout for Scenario 3, which used a ( 3  3 | 4  4 )

stope frame. As can be seen in Figure 4.6, all the stopes conformed to the minimum mining

requirement of 3 blocks in the X direction and 3 blocks in the Z direction.

Two stopes formed with the maximum dimensions demonstrating the strength of the BILP

model to generate variable and efficient stope shapes in the optimal layout. From Table

4.6, Scenario 3 recorded an economic value of $34 M by targeting and selecting 75 blocks

in total consisting of 69 ore blocks from a possible 144 and six (6) waste blocks to complete

the layout. This translates into total mineralized material of 1.1 Mt with average grade of

3.08 g/t in 1.78 hrs. As stated earlier in Section 4.2.2 the lower the grade the more metal

and thus more value. Similarly, since there are more blocks to mine, the algorithm is smart

enough to go after lower grades that meet the cutoff grade to maximize the value.

Figure 4.7 shows the optimal solution for Scenario 4, which uses a larger

( 3  5 | 4  6 ) stope frame. Similar results were achieved for this scenario relative to

Scenario 3. All the stopes in the optimal layout conformed to the minimum mining

requirement of 3 blocks in the Z direction and 5 blocks in the X direction. Also, all the

stopes formed had variable stope lengths and all respected the allowable mining

dimensions. The stopes generated had 81 blocks in total consisting of 69 ore blocks from

a possible 144 and 12 waste blocks to complete the layout. This translates into total

mineralized material of 1.2 Mt with average grade of 2.98 g/t in 1.03 hrs.
 99

The effect of adding more waste blocks is offset by the benefit of having more

oreblocks in the stopes since the algorithm does not have to leave ore behind in pillars.

This makes the stope grade positive. Also, the decision time to include a block or not is

reduced since the frame allows consideration of more blocks to be included in a stope.

4.4. EFFECT OF STOPE PILLAR DIMENSIONS (GEOTECHNICAL

REQUIREMENT)

In naturally supported underground mining operations, ground control is essential

to ensure excavation stability, worker, and equipment safety. To achieve this goal,

geomechanical engineers design pillars based on rock mass quality, stress distribution,

potential ground instability area as well as host and surrounding rock geological

characteristics (more information in Section 2.2.2.3).

Input Data for Pillar Dimensions Evaluation. These pillar dimensions (in

number of block units for this algorithm) are then included in the algorithm to generate

stopes in the final layout that are safe, operable, and profitable. The author investigates the

effect of changing the pillar dimensions and evaluates the impact on the solution. The

reader should note that advanced computational models and algorithms are currently

employed to determine the most effective arrangement of pillars within the stope layout.

Table 4.7 shows the different pillar lengths used in this experiment while Table 4.8 shows

the BILP model configuration for the three (3) scenarios in this experiment.
 100

Table 4.7 Experiment 3 – Pillar Dimensions

BILP Configuration Scenario 1 Scenario 2 Scenario 3

Minimum Pillar Length 𝛾1 2 3 2

Minimum Pillar Length 𝛾2 2 3 4

Table 4.8 BILP Input Data – Experiment 3

BILP Configuration Scenario

Minimum Mining Height 𝛼1 3

Minimum Mining Width 𝛼2 3

Maximum Mining Height 𝛽1 3

Maximum Mining Width 𝛽2 3

Number Of Stopes k 15

Cutoff Grade 𝐺𝑜𝑓𝑓 2 g/t

Results and Discussion. Table 4.9 summarizes the results while Figure 4.8

– Figure 4.10 show the optimal stope layout for the scenarios. The results show that the

objective function value drops from $27M to a low $23M as the pillar size increases in this

experiment. The findings from the study reveal that in the optimal stope layout, larger pillar

sizes result in a greater number of blocks being designated as pillars, contributing to

enhanced structural stability.

 96

Table 4.9 Experiment 3- Results of Changing Pillar Dimensions

Optimization Results
Parameter Units
Scenario 1 Scenario 2 Scenario 3
Objective Function Value ($) 27,455,739.1 24,472,902.5 23,467,732.5

Solution Time (hrs) 2.26 1.70 2.36

Gap Tolerance (%) 0.00 0.00 0.00

Number Of Stopes (#) 5 5 5

Ore Blocks Mined (#) 43 39 37

Waste Blocks Mined (#) 2 6 8

Number Of Mined Blocks (#) 45 45 45

Number Of Pillar Blocks (#) 60 75 90

Minimum Stope Grade (g/t) 2.49 2.15 2.36

Maximum Stope Grade (g/t) 5.52 5.52 5.52

Average Layout Grade (g/t) 3.86 3.53 3.42

Total Layout Tonnage (tonnes) 668,250 668,250 668,250

101
 102

Figure 4.8 Experiment 3 Scenario 1 Optimal Layout

102
 103

Figure 4.9 Experiment 3 Scenario 2 Optimal Layout

103
 104

Figure 4.10 Experiment 3 Scenario 3 Optimal Layout

104
 105

Conversely, selecting smaller pillars leads to a more compact layout, effectively

minimizing the number of blocks left behind as pillars and optimizing resource extraction.

There is no consistent trend in the solution times. All scenarios achieved the same overall

block count (45 blocks) due to the square stope frame configuration however the smaller

the pillar size the more ore blocks are mined and the larger the pillar size the more waste

blocks are selected in the stopes. This is further supported by the finding that when pillar

sizes grow, the optimal stope layout's average grade similarly falls. All these lead to the

lower objective function value as the pillar size increases in the problem.

Each scenario had pillars in units of blocks forming around the stopes in the vertical

and horizontal directions in the final layout. The pillars in the X-direction are more visible

because of the orientation of the deposit. The spatial arrangement of the stopes in the

layouts move from compact to sparse as pillar size increases. Figure 4.8 shows the result

of Scenario 1 with all the five (5) stopes formed respected the pillar dimensions constraint

with two units of blocks (pillars) separating each adjacent stope in the layout. The stopes

generated also looked compact in terms of spatial distance to each other because of

selecting a smaller dimension for the pillars between the stopes.

Figure 4.9 shows the optimal layout for Scenario 2, which contains five (5) stopes

in the optimal layout each spatially separated by 3 units of blocks in both Z and X

directions. The stopes generated are more spread out spatially in the final layout due to an

increase in the minimum pillar requirement. The algorithm selects blocks farthest from the

deposit's key central zone thus generating two (2) low-grade stopes, two (2) medium-grade

stopes and one (1) high-grade stope. Two stopes (stopes 1 and 3) lie diagonally adjacent in

the middle zone, highlighting the previously discussed limitation of this BILP model.
 106

Figure 4.10 shows the optimal layout of Scenario 3, which contains five (5) stopes

that respect the pillar dimension constraint. A total of 90 blocks were left as pillar blocks,

which resulted in the stopes formed in the final layout being widely spaced apart. This also

explains the reason why it takes a long time to converge to a solution since the algorithm

needs to find stopes that meet, the cutoff grade, the shape requirement as well as respect

the required minimum pillar dimensions.

4.5. EFFECT OF NUMBER OF STOPES SELECTED

The number of stopes K is an important parameter that controls the number of

variables and constraints created in the optimization problem. The number of stopes is

chosen a priori as an input for this algorithm.

Input Data for Number of Stopes Selected Evaluation. It is important that

the engineer avoids selecting a value of K that is lower than the number of "optimal" stopes

for the specific problem. Otherwise, the problem will converge to a suboptimal solution.

On the other hand, because K is directly related to the number of decision variables and

constraints, too large a value of K will unnecessarily increase the computational time. The

author investigated the effect of changing the number of stopes chosen for the optimization.

Table 4.10 shows the variable stope numbers selected for each scenario in this experiment

while Table 4.11 shows the BILP configuration for the experiment.
 107

Table 4.10 Experiment 4 – Number of Stopes

BILP Scenario Scenario Scenario Scenario

Configuration 1 2 3 4
Number Of Stopes k 4 10 20 30

Table 4.11 BILP Input Data – Experiment 4

BILP Configuration Scenario

Minimum Mining Height 𝛼1 3

Minimum Mining Width 𝛼2 3

Maximum Mining Height 𝛽1 3

Maximum Mining Width 𝛽2 3

Minimum Pillar Length 𝛾1 2

Minimum Pillar Length 𝛾2 2

Cutoff Grade 𝐺𝑜𝑓𝑓 2 g/t

 108

Results and Discussion. Table 4.12 shows a summary of the results while

Figure 4.11 – Figure 4.13 show the optimal layouts of the scenarios. The result for Scenario

2 in this experiment is the same as Scenario 2 of Experiment 2. The results in Table 4.12

shows the solution time increases as the number of stopes increases, as one would expect.

This is because of the explosion of variables and constraints from the higher number of

stopes. When the number of stopes was lower than the optimal number of stopes (Scenario

1), the objective function value was lower than the optimal objective function value. The

objective function value for the other scenarios (where the specified number of stopes is

higher than the optimal number of stopes) is the same for all scenarios. This is what one

would expect for this problem.

The limitation of the proposed BILP approach illustrated by this result is that the

engineer seeking to optimize his/her stope layout needs to select the maximum number of

stopes a priori that can result in an optimal solution. Otherwise, the model can yield

suboptimal results as shown in this experiment. To address this the author recommends

selecting a large enough number of stopes (e.g., by estimating the maximum number of

stopes that will fit the domain if there was to be a stope in every region possible). The

downside of this approach is that it will lead to the algorithm taking a long time to converge

to an optimal solution.
 109

Table 4.12 Experiment 4- Results of Changing Number of Stopes

Optimization Results
Parameter Units
Scenario 1 Scenario 2 Scenario 3 Scenario 4
Objective Function Value ($) 24,828,692 27,455,739 27,455,739 27,455,740

Solution Time (hrs) 0.07 0.66 3.37 9.05

Gap Tolerance (%) 0.00 0.00 0.00 0.00

Number Of Stopes (#) 4 5 5 5

Ore Blocks Mined (#) 34 43 43 43

Waste Blocks Mined (#) 2 2 2 2

Number Of Mined Blocks (#) 36 45 45 45

Number Of Pillar Blocks (#) 48 60 60 60

Minimum Stope Grade (g/t) 3.63 2.49 2.49 2.49

Maximum Stope Grade (g/t) 5.52 5.52 5.52 5.52

Average Layout Grade (g/t) 4.25 3.86 3.86 3.86

Total Layout Tonnage (tonnes) 534,600 668,250 668,250 668,250

109
 110

Figure 4.11 Experiment 4 Scenario 1 Optimal Layout

110
 111

Figure 4.12 Experiment 4 Scenario 2 Optimal Layout

111
 112

Figure 4.13 Experiment 4 Scenario 3 Optimal Layout

112
 113

Figure 4.14 Experiment 4 Scenario 4 Optimal Layout

113
 114

4.6. EFFECT OF OPTIMIZATION PROBLEM SIZE

As was previously mentioned in Section 3.8, binary integer programming models

can evolve into a combinatorial explosion of variables and constraints. The author assesses

the impact of applying the BILP model to solve different sized stope layout optimization

problems.

Input Data for Optimization Problem Size Evaluation. To conduct this

experiment the author reblocked the sample data set used for the basecase study into a two

blockmodels with smaller block dimensions consequently generating more blocks in those

blockmodels (1,000 blocks and 1,300 blocks). The author then generated economic values

for the blockmodels for this experiment using the same procedure that was employed for

the basecase scenario in Section 3.7.2.

To ensure consistency in the results, the scenarios in this experiment were solved

using the same MATLAB code used in the prior analysis. The complete input data for this

experiment can be accessed on GitHub (https://github.com/TheoMensah/BILP_SSLO.git).

Table 4.13 shows the summary statistics of the block models. Table 4.14 shows the

common input data for this experiment while Table 4.15 shows the number of blocks for

each of the three models used in this experiment.

Results and Discussion. Table 4.16 shows a summary of the results of this

experiment and Figure 4.16 –Figure 4.18 show the layouts. The result from this study

suggests that the size of the optimization problem does have an impact on the solution

obtained and this is evident in the solutions times achieved in each scenario (Figure 4.15).

All formulations converged to an optimal solution with gap tolerance of 0%.

 115

Table 4.13 Summary Statistics of Block Models

Blockmodel Reblocked Reblocked Reblocked

Attribute Model 1 Model 2 Model 3
Metal Au Au Au

Number of blocks (#) 774 1,000 1,300

Blocks Au > 0 144 191 246

Total mineralized material (t) 2,138,400 2,138,400 2,138,400

Maximum Au value (g/t) 14.91 9.72 9.97

Minimum Au value (g/t) 0.15 0.11 0.29

Average Au value (g/t) 2.63 2.55 2.59

Standard deviation (%) 2.09 1.66 1.67

Variance (%2) 4.35 2.74 2.80

Block density (kg/m3) 2.2 2.2 2.2

Block size (#) 15 ×15 ×30 13 ×15 ×27 10 ×15 ×27

Block tonnage (t) 6,750 5,225 4,050

Depth from surface (m) 560 – 1,100 560 – 1,100 560 – 1,100
 116

Table 4.14 BILP Input Data – Experiment 5

BILP Configuration Scenario

Minimum Mining Height 𝛼1 4

Minimum Mining Width 𝛼2 4

Maximum Mining Height 𝛽1 4

Maximum Mining Width 𝛽2 4

Minimum Pillar Length 𝛾1 3

Minimum Pillar Length 𝛾2 3

Number Of Stopes k 20

Cutoff Grade 𝐺𝑜𝑓𝑓 3g/t

Table 4.15 Experiment 5 –Optimization Problem Size

BILP Configuration Scenario 1 Scenario 2 Scenario 3

Number of Blocks (i, j ) 774 1000 1300

 117

8.00

7.00

6.00

5.00
Solution Time (Hrs)

4.00
774
1000
1300
3.00

2.00

1.00

0.00
700 800 900 1000 1100 1200 1300 1400
Number of Blocks (𝑖,𝑗)

Figure 4.15 Solution Time versus Optimization Problem Size (Number of Blocks)
 118

Table 4.16 Experiment 5- Results of Changing Optimization Problem Size

Optimization Results
Parameter Units
Scenario 1 Scenario 2 Scenario 3
Objective Function Value ($) 24,627,142.8 24,990,276.6 25,358,764.9

Solution Time (hrs) 1.28 3.83 7.37

Gap Tolerance (%) 0.00 0.00 0.00

Number of Stopes formed (#) 3 3 5

Ore Blocks Mined (#) 46 46 78

Waste Blocks Mined (#) 2 0 2

Number of mined blocks (#) 48 48 80

Number of pillar blocks (#) 72 95 130

Minimum Stope Grade (g/t) 3.08 3.22 3.12

Maximum Stope Grade (g/t) 3.92 3.55 3.70

Average Layout Grade (g/t) 3.38 3.37 3.45

Total Layout Tonnage (tonnes) 712,800 712,800 712,800

118
 119

Figure 4.16 Experiment 5 Scenario 1 Optimal Layout

119
 120

Figure 4.17 Experiment 5 Scenario 2 Optimal Layout

120
 121

Figure 4.18 Experiment 5 Scenario 3 Optimal Layout

121
 122

From the study, Figure 4.15 illustrates a scatter plot between optimization problem

(number of blocks) and the solution time achieved. The plot demonstrates that there is a

linear correlation between the optimization problem size and the time required to converge

to an optimal solution. This relationship occurs because as the number of blocks increases,

the decision variables and constraints also grow, resulting in a more complex problem

necessitating greater computational time and resources for the algorithm to converge

towards an optimal solution. The results in Table 4.16 indicates Scenario 1 (blocks (i, j ) =

774) had the fastest solution time of 1.28 hrs. Scenario 2 (blocks (i, j ) = 1,000) achieved

this in 3.83 hrs while Scenario 3 (blocks (i, j ) = 1,300) achieved the optimal solution in the

longest time of 7.37 hrs. While this experiment shows a linear relationship, it is not yet

clear whether this is the case for a broad range of problem sizes. If indeed, the

computational time grows only linearly, this will be an advantage of this model. Further

work is required to examine this observation.

This findings from this experiment demonstrates yet another flaw in the BILP

approach, namely the tendency for variable and constraint combinations to explode as

optimization problem size increases requiring greater computational time and resources for

the algorithm to converge to an optimal solution.

From Figure 4.16 – Figure 4.18 the author reblocked the basecase model into

varying block sizes. To maintain consistency in this experiment, the size of the pillars and

stopes dimensions were kept the same. However due to the varying block sizes the final

designs will change. This approach was adopted to avoid introducing additional

confounding factors, such as changes in the number of pillar and stope constraints per block
 123

that would occur with different stope sizes. The reblocking process redistributes the grade

in the model as well as increase the block count for small sized blocks.

From the results, smaller block sizes allow the algorithm more selectivity. Thus, as

can be seen in the optimal layout of Scenario 3 (Figure 4.18), the algorithm had the

flexibility to include more blocks hence forming 5 stopes – 2 more than Scenarios 1 and 2.

This is because the smaller sized blocks have a lower tonnage and to achieve the maximum

value more higher grade stopes needs to be formed in order to achieve the maximum

economic value. Though the same final layout tonnage is achieved in all scenarios,

objective function value is slightly higher for the smaller block sized problem (Scenario

3). This is because grade variability is also increased in the smaller block sizes making the

algorithm more selective in the blocks to include in a stope.

4.7. SUMMARY

This section presented a set of computational experiments to assess the proposed

BILP model's sensitivity to the key input parameters of the stope layout optimization

problem. The work in this chapter evaluated the effect of differences in cutoff grade, stope

dimensions, pillar dimensions, (maximum) number of stopes selected by user, and size of

the optimization problem (measured in terms of the number of blocks).

The following are the major findings from the experiments:

• The BILP model can find the optimal solution for many different types of

problems. For all the scenarios evaluated, the solution was found within an

optimality gap of 0.00%.

 124

• The model is sensitive to changes in the cutoff grade. While a high cutoff grade

will speed up the algorithm's solution time, this will generate optimal layouts

with a lower objective function value.

• The model can generate stopes of any rectangular shape (stope frame) specified

by the engineer and will mimic the deposits peripheries. The stope frames the

engineer selects affects the objective function value, solution time, number of

stopes formed, and general stope layout.

• The model is also sensitive to the specified pillar sizes. Specifying larger pillar

sizes results in more spatially spread stopes in the layout which leads to optimal

layouts with a lower objective function (economic) value. Specifying smaller

pillars lead to more compact stopes in the optimal layout with higher objective

function (economic) values.

• The model is highly sensitive to specified (maximum) number of stopes and the

number of blocks (used as a proxy for optimization problem size) because both

parameters affect the number of decision variables and constraints. The larger

the specified (maximum) number of stopes and the higher the number of blocks

the longer the solution time.

• Because the engineer needs to select the maximum number of stopes a priori

and it has such a significant effect on solution times and the solution, this work

proposes that engineers using this model estimate the maximum number of

stopes possible in the geometry and use that estimate for the maximum number

of stopes. This ensures there are enough stopes to yield the optimum solution

but not more than necessary.

 125

5. CONCLUSIONS, RECOMMENDATIONS & FUTURE WORK

5.1. OVERVIEW

Stope layout optimization is a critical aspect of underground mining operations. It

involves determining the most effective arrangement of stopes within a mine to maximize

resource extraction and operational efficiency. The primary objective is to design a layout

that optimizes the economic value while considering various factors such as geotechnical

constraints, stope grade, equipment limitations, and safety requirements.

Previous researchers have used meta-heuristic optimization methods, including

swarm intelligence algorithms, genetic algorithms, and particle swarm optimization, to find

optimal stope layouts. These approaches however do not guarantee optimality. One

approach that researchers are utilizing now involves formulating the problem as a

mathematical optimization model, typically using linear programming techniques to

determine the optimal arrangement of stopes. This approach is well known to guarantee

optimality and if well formulated, can be configured to solve complex problems.

Thus, the goal of this thesis work was to formulate the stope layout optimization

problem (SLOP) as a binary integer linear programming problem that maximizes the value

of the stopes mined subject to novel grade, geotechnical (minimum and maximum pillar

sizes), and allowable mining (minimum and maximum stope width and height) constraints

in two-dimensional space. To achieve this goal, the author:

1. Drew from Queyranne and Wolsey’s [25], [26] formulations of tight

constraints for bounded up/down times in production planning problems to formulate novel
 126

and efficient geometric constraints along with geotechnical and grade constraints for the

BILP stope layout optimization problem.

2. verified the novel BILP model with a sample gold mining data set to verify

the model. The original geological model of the orebody was regularized to generate equal-

sized blocks ideal for conversion into an economic model which served as the primary

input for the 5-experimental 15-scenario runs to verify the BILP model as a model that

applies efficient shape constraints in solving the SLOP in two-dimensional space.

5.2. CONCLUSIONS

The study concludes the following from the outcome of the basecase and

experimental computations:

▪ With respect to the outcomes from the basecase:

❖ The results from the basecase study show that it is possible to model

shapes using LP-based techniques for the stope layout problem. Unlike

most LP-based stope layout approaches, the proposed model accounts

for efficient shape constraints in the geometric constraints.

❖ The developed model can find the optimal stope layout that maximizes

the undiscounted profit for the deposit within a gap tolerance of 0.00%.

❖ The model allows the user to generate variable stope length and height

as well as incorporate geotechnical pillar requirements between the

generated stopes. Thus, the model permits mining operations to follow

irregular mineral deposit peripheries to minimize dilution while

ensuring a stable operating environment.

 127

❖ The proposed BILP model has some limitations as illustrated by the

results of the base case experiment. Pillars are respected around the

stopes however pillar widths are not maintained in the diagonal

direction because the pillar limitations are defined along the vertical and

horizontal directions. This will require future post-processing to ensure

the stope layouts are safe.

▪ With respect to the computational experiments to evaluate the BILP model’s

sensitivity to key input variables and parameters:

❖ The proposed BILP model is sensitive to the selected stope cutoff grade.

The lower the cutoff grade selected the higher the objective function

achieved and vice versa. Also, there is a trade-off between the solution

time and the objective function value achieved. A high stope cutoff

grade means solution converges faster but it does not improve the

objective function value.

❖ The model is also very sensitive to the required minimum stope

dimensions. As stated above, the model can generate variable stope

frames based on the input. Larger stope frames achieve higher objective

function values (assuming pillar sizes stay the same) and converges

faster since more blocks can be selected and fewer blocks are left behind

as pillars. Smaller frames generate more stopes in the layout but requires

more pillars thus leaving some blocks behind and achieving a lower

objective function value.

 128

❖ The model is sensitive to the specified pillar dimensions. The smaller

the pillar dimensions selected the fewer the blocks left behind as pillars

thus maximizing the economic value of the layout and vice versa. The

solution time is not affected by the choice of pillar.

❖ The performance of the model is affected by the choice of maximum

number of stopes. The maximum number of stopes is chosen a priori

which means the engineer can select a larger number which will lead to

a longer time for the algorithm to converge or a smaller number which

will lead to suboptimal solutions. The author proposes that engineers

using this model estimate the maximum number of stopes possible in

the geometry and use that estimate for the maximum number of stopes.

This ensures there are enough stopes to yield the optimum solution but

not more than necessary.

❖ The BILP model proposed is highly sensitive to the size of the

optimization problem. There appears to be a linear correlation between

the problem size and the solution time. The larger the problem size the

more exponentially the variables and constraints grow making the

problem more complex and necessitating greater computational

resources and time to solve such problems. However, because the

experiments in this work are very limited, it is not clear if the trend is

the same for a wide variety of problem sizes.

▪ The proposed model contributes to the research on underground stope layout

optimization by demonstrating the possibility to formulate efficient shape

 129

constraints in a binary integer programming model to solve the stope layout

optimization problem. The work evaluated systematically, the sensitivity and

performance of the proposed BILP model with respect to stope layout

optimization’s key input parameters as well as optimization problem size. The

work also demonstrated the adaptation of Queyranne and Wolsey’s [25], [26]

tighter constraints for production planning to model novel tighter and more

efficient constraints for the stope layout optimization problem.

5.3. RECOMMENDATIONS FOR FUTURE WORK

To further improve and advance the proposed binary-integer linear programming

model for optimizing the stope layout optimization problem, the following

recommendations are made for future work:

❖ The model currently does not implement pillars diagonally which means stopes

can be generated diagonally adjacent to each other. This can lead to stability

issues during stoping. Future work should develop post-processing algorithms

that can detect “diagonal” pillars that violate pillar constraints and use heuristics

to adjust the layout to avoid these situations.

❖ The utilization of binary variables in this model leads to longer solution times,

necessitating the implementation of strategies to reduce the computational cost

associated with the model. One such strategy is to preprocess the problem using

heuristic techniques prior to passing the problem to a standard BILP solver.

Future work should develop such preprocessing algorithms to improve the

solution times.
 130

❖ Future work should account for mine access networks in the solution. Mine

development layouts are essential for stope layout design and should be

incorporated into the stope layout optimization problem.

❖ It will be essential to integrate the stope production scheduling problem with

this model.

❖ Due to the high uncertainty that characterizes mining operations, it is

recommended that future work should model a stochastic optimization model

of this BILP model to address uncertainty in model parameters.

❖ Lastly, the proposed model is in two-dimensional space (2D). To make this

model more practical with realistic outputs for application in real-life mining

scenarios, the model should be extended to three-dimensional (3D) space.

Future work should incorporate a third set of variables and constraints in the

third dimension to extend the proposed model’s framework to 3D space.

 131

BIBLIOGRAPHY

[1] A. Copco, “Mining methods in underground mining,” in Atlas Copco: Nacka,

Sweden, 2007.

[2] P. Darling, SME Mining Engineering Handbook, Third Edition., vol. 1.

Englewood, Co: Society for Mining, Metallurgy and Exploration, 2011.

[3] S. Mijalkovski, D. Peltechki, Z. Despodov, D. Mirakovski, V. Adjiski, and N.

Doneva, “METHODOLOGY FOR UNDERGROUND MINING METHOD
SELECTION,” Mining Science, vol. 28, pp. 201–216, 2021, doi:
10.37190/msc202815.

[4] F. e F. Matheus, R. Dimitrakopoulos, and C. Pinto, “Stochastic stope design

optimisation under grade uncertainty and dynamic development costs,” Int J Min
Reclam Environ, vol. 36, no. 2, pp. 81–103, 2022, doi:
https://doi.org/10.1080/17480930.2021.1968707.

[5] G. Samanta, T. Dey, and B. Samanta, “Stope boundary optimization for an

underground copper deposit using mixed integer linear programming based
algorithm,” Journal of Mines, Metals and Fuels, vol. 69, p. 73, Jan. 2021, doi:
10.18311/jmmf/2021/27781.

[6] Y. A. Sari and M. Kumral, “Sublevel stope layout planning through a greedy
heuristic approach based on dynamic programming,” Journal of the Operational
Research Society, vol. 72, no. 3, pp. 554–563, 2021, doi:
10.1080/01605682.2019.1700179.

[7] D. S. S. Sandanayake, E. Topal, and M. W. A. Asad, “Designing an optimal stope

layout for underground mining based on a heuristic algorithm,” Int J Min Sci
Technol, vol. 25, no. 5, pp. 767–772, 2015, doi:
https://doi.org/10.1016/j.ijmst.2015.07.011.

[8] G. Erdogan, M. Cigla, E. Topal, and M. Yavuz, “Implementation and comparison

of four stope boundary optimization algorithms in an existing underground mine,”
Int J Min Reclam Environ, vol. 31, no. 6, pp. 389–403, Aug. 2017, doi:
10.1080/17480930.2017.1331083.
 132

[9] G. Erdogan, M. Cigla, E. Topal, and M. Yavuz, “Implementation and comparison

of four stope boundary optimization algorithms in an existing underground mine,”
Int J Min Reclam Environ, vol. 31, no. 6, pp. 389–403, Aug. 2017, doi:
10.1080/17480930.2017.1331083.

[10] B. O. Afum and E. Ben-Awuah, “A Review of Models and Algorithms for

Surface-Underground Mining Options and Transitions Optimization: Some
Lessons Learnt and the Way Forward,” Mining, vol. 1, no. 1, pp. 112–134, May
2021, doi: 10.3390/mining1010008.

[11] K. Shahriar and K. Oraee, “A Study on the Optimization Algorithms for

Determining Open-Pit and Underground Mining Limits,” in 7th International
Scientific Conference, SGEM 2007, International Multidisciplinary Scientific
GeoConferences (SGEM), 2007.

[12] M. Osanloo, J. Gholamnejad, and B. Karimi, “Long-term open pit mine production
planning: A review of models and algorithms,” International Journal of Mining,
Reclamation and Environment, vol. 22, no. 1. pp. 3–35, Mar. 2008. doi:
10.1080/17480930601118947.

[13] E. J. A. Appianing, E. Ben-Awuah, and Y. Pourrahimian, “Life-of-mine

optimization for integrated open stope development and production scheduling
using a mixed-integer linear programming framework,” Mining Technology, pp. 1–
15, Mar. 2023, doi: 10.1080/25726668.2023.2182285.

[14] J. Little and E. Topal, “Strategies to assist in obtaining an optimal solution for an
underground mine planning problem using Mixed Integer Programming,” Int J
Min Miner Eng, vol. 3, no. 2, pp. 152–172, 2011.

[15] J. Manchuk and C. Deutsch, “Optimizing stope designs and sequences in

underground mines,” SME Transactions, vol. 324, pp. 67–75, Jan. 2008.

[16] J. Little, E. Topal, and P. Knights, “Simultaneous optimisation of stope layouts and
long term production schedules,” Transactions of the Institutions of Mining and
Metallurgy, Section A: Mining Technology, vol. 120, no. 3, pp. 129–136, 2011,
doi: 10.1179/1743286311Y.0000000011.

[17] J. Hou, C. Xu, P. A. Dowd, and G. Li, “Integrated optimisation of stope boundary
and access layout for underground mining operations,” Mining Technology, vol.
128, no. 4, pp. 193–205, 2019, doi: 10.1080/25726668.2019.1603920.
 133

[18] A. Chowdu, P. Nesbitt, A. Brickey, and A. M. Newman, “Operations Research in

Underground Mine Planning: A Review,” Interfaces (Providence), vol. 52, no. 2,
pp. 109–132, 2022, doi: 10.1287/inte.2021.1087.

[19] E. J. A. Appianing and E. Ben-Awuah, “Underground Mining Stope Layout

Optimization and Production Scheduling: A Review of Existing Solvers and
Algorithms,” MOL Report Nine, pp. 271–304, 2018.

[20] A. S. Nhleko, T. Tholana, and P. N. Neingo, “A review of underground stope

boundary optimization algorithms,” Resources Policy, vol. 56, pp. 59–69, Jun.
2018, doi: 10.1016/j.resourpol.2017.12.004.

[21] V. Nikbin, E. Mardaneh, M. W. A. Asad, and E. Topal, “Pattern search method for
accelerating Stope boundary optimization problem in underground mining
operations,” Taylor and Francis Ltd., 2022. doi:
10.1080/0305215X.2021.1932869.

[22] S. A. Abdulsalaam, “A FAST DYNAMIC PROGRAMMING ALGORITHM FOR

STOPE BOUNDARY LAYOUT FOR UNDERGROUND MINE,” Int J Adv Res
(Indore), vol. 9, no. 09, pp. 86–90, Sep. 2021, doi: 10.21474/IJAR01/13378.

[23] S. E. Jalali and M. Ataee-pour, “A 2D dynamic programming algorithm to

optimize stope boundaries,” in Mine Planning and Equipment Selection 2004:
Proceedings of the Thirteenth International Symposium on Mine Planning and
Equipment Selection, Wroclaw, Poland, 1-3 September 2004, 2004, p. 45.

[24] Z. Basiri and Y. Pourrahimian, “Application of mathematical programming for

stope layout and production schedule optimization in sublevel stoping,” Mining
Optim. Laboratory, vol. 8, no. 1, pp. 235–245, 2018.

[25] M. Queyranne and L. A. Wolsey, “Tight MIP formulations for bounded up/down
times and interval-dependent start-ups,” Math Program, vol. 164, no. 1–2, pp.
129–155, Jul. 2017, doi: 10.1007/s10107-016-1079-2.

[26] M. Queyranne, “Extended Formulations for Shape Constraints in Mine Planning,”

in DEPLAMIN, Santa Cruz Chile, 2018.
 134

[27] J. Fuykschot, “Strategic Mine Planning Flexible mine planning to meet changes in
the business environment,” in Minex Conference 2009, SRK Consulting, 2009, pp.
1–18.

[28] M. Emelly and V. Matamoros, “A Novel Stope Layout Optimization Based on

Robust Genetic Algorithms,” Doctoral Thesis, McGill University (Canada), 2018.

[29] D. Suneth and S. Sandanayake, “Stope Boundary Optimization in Underground

Mining Based on a Heuristic Approach,” Doctoral Thesis, Curtin University, 2014.

[30] A. M. Newman, E. Rubio, R. Caro, A. Weintraub, and K. Eurek, “A review of

operations research in mine planning,” Interfaces (Providence), vol. 40, no. 3, pp.
222–245, May 2010, doi: 10.1287/inte.1090.0492.

[31] J. De la Vergne, “Hard Rock Miners Handbook, edition 5,” Ground Water, vol. 15,
pp. 35–45, 2008.

[32] C. Musingwini, “Presidential address: Optimization in underground mine

planning-developments and opportunities,” J South Afr Inst Min Metall, vol. 116,
no. 9, pp. 809–820, Sep. 2016, doi: 10.17159/2411-9717/2016/v116n9a1.

[33] M. Janiszewski, S. Pontow, and M. Rinne, “Industry Survey on the Current State
of Stope Design Methods in the Underground Mining Sector,” Energies (Basel),
vol. 15, no. 1, Jan. 2022, doi: 10.3390/en15010240.

[34] “Review of:‘Techniques in Underground Mining: Selections from Underground

Mining Methods Handbook’ (1998). Edited by Richard E. Gertsch and Richard L.
Bullock. Society for Mining, Metallurgy, and Exploration Inc. P.O. Box 625002.
Littleton, CO, U.S.A. 80162-5002.,” International Journal of Surface Mining,
Reclamation and Environment, vol. 13, no. 1, pp. vii–vii, Jan. 1999, doi:
10.1080/09208119908944193.

[35] H. Hamrin, “Choosing an underground mining method,” in Underground mining

methods handbook, W. A. Hustrulid, Ed., New York: AIME, 1982.

[36] P. Darling, “SME Underground Mining Handbook,” vol. 1. Society for Mining,
Metallurgy, and Exploration (SME), 2023. [Online]. Available:
https://app.knovel.com/hotlink/toc/id:kpSMEUMH01/sme-underground-
mining/sme-underground-mining
 135

[37] H. Hamrin, “Underground mining methods and applications,” in Underground

mining methods: engineering fundamentals and international case studies, W. A.
Hustrulid and R. L. Bullock, Eds., Littleton: Society for Mining Metallurgy and
Exploration, 2001.

[38] H. Harraz, “Sublevel Stoping Underground Mining Methods,” 2014.

https://www.researchgate.net/publication/301834296_Sublevel_Stoping_Undergro
und_Mining_Methods (accessed Jun. 10, 2023).

[39] S. Xu, R. Liang, F. T. Suorineni, and Y. Li, “Evaluation of the use of sublevel
open stoping in the mining of moderately dipping medium-thick orebodies,” Int J
Min Sci Technol, vol. 31, no. 2, pp. 333–346, 2021.

[40] N. Morales Varela, D. Mancilla, R. Miranda, and J. Vallejos, “Optimal Drift and
Level Design in Underground Mining Extracted by Sublevel Open Stoping
Method,” SSRN Electronic Journal, 2021, doi: 10.2139/ssrn.3928373.

[41] G. Samanta, S. Sinha, and T. Dey, “A Case Study on Economic Impact

Assessment of Developing Single Drill Level in Sublevel Stoping Method,”
Journal of The Institution of Engineers (India): Series D, 2023, doi:
10.1007/s40033-023-00453-2.

[42] B. H. G. Brady and E. T. Brown, “Rock mechanics and mining engineering,” Rock
Mechanics for underground mining: Third edition, pp. 1–16, 2006, doi:
https://doi.org/10.1007/978-1-4020-2116-9.

[43] K. Kaklis, Z. Agioutantis, M. Masialeti, J. Yendaw, and T. B. Betsi, “Parametric

Analysis of Rib Pillar Stability in a Longitudinal Sublevel Open Stoping Operation
in an Underground Copper Mine in Southern Africa,” MDPI AG, Nov. 2021, p.
11. doi: 10.3390/materproc2021005011.

[44] Y. A. Sari and M. Kumral, “A planning approach for polymetallic mines using a
sublevel stoping technique with pillars and ultimate stope limits,” Engineering
Optimization, vol. 52, no. 6, pp. 932–944, Jun. 2020, doi:
10.1080/0305215X.2019.1624739.

[45] E. S. Villaescusa, Geotechnical Design for Sublevel Open Stoping, 1st ed. CRC
Press, 2014. doi: 10.1201/b16702.
 136

[46] A. S. Nhleko and C. Musingwini, “Optimisation of Three-Dimensional Stope

Layouts Using a Dual Interchange Algorithm for Improved Value Creation,”
Minerals, vol. 12, no. 5, May 2022, doi: 10.3390/min12050501.

[47] M. T. Bootsma, C. Alford, J. Benndorf, and M. W. N. Buxton, “Cut-off grade

based sublevel stope mine optimisation: Introduction and evaluation of an
optimisation approach and method for grade risk quantification,” in Advances in
Applied Strategic Mine Planning, Springer International Publishing, 2018, pp.
537–557. doi: 10.1007/978-3-319-69320-0_31.

[48] J. J. Sens and E. Topal, “A new algorithm for Stope boundary optimization,” in
The AusIMM Newe Leaders’ Conference, Brisbane, QLD, Australia, 2009, pp. 1–
4.

[49] J. Hou, C. Xu, P. A. Dowd, and G. Li, “Integrated optimisation of stope boundary
and access layout for underground mining operations,” Mining Technology:
Transactions of the Institute of Mining and Metallurgy, vol. 128, no. 4, pp. 193–
205, Oct. 2019, doi: 10.1080/25726668.2019.1603920.

[50] C. Alford, “Optimization in underground mine design,” in 25th International

Application of Computers and Operations Research in the Mineral Industry
symposium, Brisbane, Australia: AusIMM, 1995, pp. 21–31.

[51] F. Sotoudeh, R. Kakaie, and M. Ataei, “Development of a computer program for

underground mine stope optimisation using a heuristic algorithm,” Australian
Centre for Geomechanics, 2017, pp. 689–700. doi:
10.36487/acg_rep/1710_58_soutoudeh.

[52] N. Castanho, “Mineable Stope Optimiser (MSO): A Mine Planning Tool More
Important Than Ever,” SRK Publications, Sep. 2020.

[53] M. Ataee-pour, “A Critical Survey of the Existing Stope Layout Optimization

Techniques,” Journal of Mining Science, vol. 41, no. 5, pp. 447–466, 2005, doi:
10.1007/s10913-006-0008-9.

[54] A. D. Mwangi, Z. Jianhua, H. Gang, R. M. Kasomo, and M. M. Innocent,

“Ultimate Pit Limit Optimization Methods in Open Pit Mines:A Review,” Journal
of Mining Science, vol. 56, no. 4, pp. 588–602, Jul. 2020, doi:
10.1134/S1062739120046885.
 137

[55] J. Ovanic and D. S. Young, “Economic Optimization of Open Stope Geometry,” in

APCOM -INTERNATIONAL SYMPOSIUM-,28th, Computer applications in the
minerals industries, Colorado School of Mines; , 1999, pp. 855–862. [Online].
Available: https://www.tib.eu/de/suchen/id/BLCP%3ACN035562679

[56] M. Ataee-Pour, “A heuristic algorithm to optimise stope boundaries,” 2000.

[Online]. Available: http://ro.uow.edu.au/theses/2923

[57] A. S. Nhleko, T. Tholana, and P. N. Neingo, “A review of underground stope

boundary optimization algorithms,” Resources Policy, vol. 56, pp. 59–69, 2018,
doi: https://doi.org/10.1016/j.resourpol.2017.12.004.

[58] E. Topal and J. Sens, “A new algorithm for stope boundary optimization,” Journal
of Coal Science and Engineering, vol. 16, no. 2, pp. 113–119, Jun. 2010, doi:
10.1007/s12404-010-0201-y.

[59] D. S. S. Sandanayake, E. Topal, and M. W. A. Asad, “Designing an optimal stope

layout for underground mining based on a heuristic algorithm,” Int J Min Sci
Technol, vol. 25, no. 5, pp. 767–772, Sep. 2015, doi: 10.1016/j.ijmst.2015.07.011.

[60] M. E. V. Matamoros and M. Kumral, “Heuristic stope layout optimisation

accounting for variable stope dimensions and dilution management,” Int J Min
Miner Eng, vol. 8, no. 1, pp. 1–18, 2017.

[61] V. Nikbin, M. Ataee-pour, K. Shahriar, and Y. Pourrahimian, “A Greedy

Algorithm for Stope Boundaries Optimization,” Mining Optimization Laboratory
(MOL) , vol. 8, pp. 246–252, 2017.

[62] T. M. Mmola, A. S. Nhleko, and J. M. Atherfold, “Application of Particle Swarm

Optimization Algorithm to Optimize Stope Layout for Underground Mines,” in
Proceedings of the 27th International Symposium on Mine Planning and
Equipment Selection - MPES 2018, E. Widzyk-Capehart, A. Hekmat, and R.
Singhal, Eds., Cham: Springer International Publishing, 2019, pp. 213–221.

[63] Y. A. Sari and M. Kumral, “Clustering-based iterative approach to stope layout

optimization for sublevel stoping,” The Journal of the Southern African Institute of
Mining and Metallurgy, vol. 121, 2021, doi: 10.17159/2411.
 138

[64] J. M. Riddle, “A Dynamic Programming Solution of A Block-Caving Mine

Layout,” in 14th Application Of Computer Methods In The Mineral Industry, 1977,
pp. 767–780.

[65] E. J. A. Appianing, E. John, A. Appianing, and E. Ben-Awuah, “Underground

Mining Stope Layout Optimization and Production Scheduling A Review of
Existing Solvers and Algorithms,” 2018.

[66] N. Grieco and R. Dimitrakopoulos, “Managing grade risk in stope design

optimisation: Probabilistic mathema-tical programming model and application in
sublevel stoping,” Min. Technology, vol. 116, no. 2, pp. 49–57, 2007, doi:
doi:10.1179/174328607X191038.

[67] X. Bai, D. Marcotte, and R. Simon, “Underground stope optimization with

network flow method,” Comput Geosci, vol. 52, pp. 361–371, 2013, doi:
10.1016/j.cageo.2012.10.019.

[68] R. W. Eglese, “Simulated annealing: A tool for operational research,” Eur J Oper
Res, vol. 46, no. 3, pp. 271–281, 1990, doi: https://doi.org/10.1016/0377-
2217(90)90001-R.

[69] T. D. Mavridou and P. M. Pardalos, “Simulated Annealing and Genetic

Algorithms for the Facility Layout Problem: A Survey,” Kluwer Academic
Publishers, 1997.

[70] M. E. Villalba Matamoros and M. Kumral, “Underground mine planning: stope

layout optimisation under grade uncertainty using genetic algorithms,” Int J Min
Reclam Environ, vol. 33, no. 5, pp. 353–370, Jul. 2019, doi:
10.1080/17480930.2018.1486692.

[71] R. L. A. Verhoeff, M. W. N. Buxton, T. U. Delft, A. M. Msc, and R. Aachen,

“Using Genetic Algorithms for Underground Stope Design Optimization in
Mining A Stochastic Analysis,” 2017. [Online]. Available:
http://repository.tudelft.nl/.

[72] J. Hou, G. Li, H. Wang, and N. Hu, “Genetic algorithm to simultaneously optimise
stope sequencing and equipment dispatching in underground short-term mine
planning under time uncertainty,” Int J Min Reclam Environ, vol. 34, no. 5, pp.
307–325, May 2020, doi: 10.1080/17480930.2019.1584952.
 139

[73] G. Nelis, M. Gamache, D. Marcotte, and X. Bai, “Stope optimization with vertical
convexity constraints,” Optimization and Engineering, vol. 17, no. 4, pp. 813–832,
Dec. 2016, doi: 10.1007/s11081-016-9321-6.

[74] X. Bai, D. Marcotte, and R. Simon, “A heuristic sublevel stope optimizer with
multiple raises,” Journal of the Southern African Institute of Mining and
Metallurgy, , vol. 114, no. 6, pp. 472–434, 2014.

[75] J. Huchette and J. P. Vielma, “A geometric way to build strong mixed-integer

programming formulations,” Operations Research Letters, vol. 47, no. 6, 2019,
doi: 10.1016/j.orl.2019.10.003.

[76] B. O. Afum, E. Ben-Awuah, and H. Askari-Nasab, “A mixed integer linear

programming framework for optimising the extraction strategy of open pit –
underground mining options and transitions,” Int J Min Reclam Environ, vol. 34,
no. 10, pp. 700–724, Nov. 2020, doi: 10.1080/17480930.2019.1701968.

[77] H. Askari-Nasab, Y. Pourrahimian, E. Ben-Awuah, and S. Kalantari, “Mixed

integer linear programming formulations for open pit production scheduling,”
Journal of Mining Science, vol. 47, no. 3, p. 338, 2011.

[78] Y. Moysoglou, “Limitations of linear programming as a model of approximate

computation,” Ph. D. Disssertation, National and Kapodistrian University of
Athens (EKPA), 2015.

[79] B. Alhijawi and A. Awajan, “Genetic algorithms: theory, genetic operators,

solutions, and applications,” Evolutionary Intelligence. Springer Science and
Business Media Deutschland GmbH, 2023. doi: 10.1007/s12065-023-00822-6.

[80] T. M. Shami, A. A. El-Saleh, M. Alswaitti, Q. Al-Tashi, M. A. Summakieh, and S.

Mirjalili, “Particle Swarm Optimization: A Comprehensive Survey,” IEEE Access,
vol. 10, pp. 10031–10061, 2022, doi: 10.1109/ACCESS.2022.3142859.

[81] S. Y. Amponsah, P. M. Takouda, and E. Ben-Awuah, “A Multiple Objective

Genetic Algorithm Approach for Stochastic Open Pit Production Scheduling
Optimisation,” Int J Min Reclam Environ, pp. 1–28, Apr. 2023, doi:
10.1080/17480930.2023.2196918.
 140

[82] O. Khare, S. Ahmed, and Y. Singh, “An Overview of Swarm Intelligence-Based

Algorithms,” in Design and Applications of Nature Inspired Optimization:
Contribution of Women Leaders in the Field, D. Singh, V. Garg, and K. Deep,
Eds., Cham: Springer International Publishing, 2022, pp. 1–18. doi: 10.1007/978-
3-031-17929-7_1.

[83] Y. A. Sari and M. Kumral, “Dig-limits optimization through mixed-integer linear

programming in open-pit mines,” Journal of the Operational Research Society,
vol. 69, no. 2, pp. 171–182, Feb. 2018, doi: 10.1057/s41274-017-0201-z.

[84] S. Kirkpatrick, C. D. Gelatt, and M. Vecchi, “Optimization by Simulated

Annealing,” Science (1979), vol. 220, pp. 671–680, 1983.

[85] S. Kirkpatrick, C. D. Gelatt, and M. Vecchi, “Optimization by Simulated

Annealing,” Science (1979), vol. 220, pp. 671–680, 1983.

[86] S. H. Zhan, J. Lin, Z. J. Zhang, and Y. W. Zhong, “List-Based Simulated

Annealing Algorithm for Traveling Salesman Problem,” Comput Intell Neurosci,
vol. 2016, 2016, doi: 10.1155/2016/1712630.

[87] P. P. Tambe, “Selective Maintenance Optimization of a Multi-component System

based on Simulated Annealing Algorithm,” in Procedia Computer Science,
Elsevier B.V., 2022, pp. 1412–1421. doi: 10.1016/j.procs.2022.01.342.

[88] A. G. Nikolaev and S. H. Jacobson, “Simulated Annealing,” 2010, pp. 1–39. doi:
10.1007/978-1-4419-1665-5_1.

[89] K. Amine, “Multiobjective Simulated Annealing: Principles and Algorithm

Variants,” Advances in Operations Research, vol. 2019, 2019, doi:
10.1155/2019/8134674.

[90] İ. İLHAN, “An improved simulated annealing algorithm with crossover operator
for capacitated vehicle routing problem,” Swarm Evol Comput, vol. 64, p. 100911,
Jun. 2021, doi: 10.1016/J.SWEVO.2021.100911.

[91] A. Y. Zomaya, “Natural and simulated annealing,” Comput Sci Eng, vol. 3, no. 6,
pp. 97–99, 2001, doi: 10.1109/MCISE.2001.963434.
 141

[92] M. E. Villalba Matamoros and M. Kumral, “Underground mine planning: stope

layout optimisation under grade uncertainty using genetic algorithms,” Int J Min
Reclam Environ, vol. 33, no. 5, pp. 353–370, Jul. 2019, doi:
10.1080/17480930.2018.1486692.

[93] A. J. Fofanah, H. R. Bundu, and J. G. Kargbo, “generic heart diseases prediction

and application of genetic algorithms in healthcare systems,” Int J Health Sci
(Qassim), 2022, doi: 10.53730/ijhs.v6ns3.9024.

[94] M. S. Umam, M. Mustafid, and S. Suryono, “A hybrid genetic algorithm and tabu
search for minimizing makespan in flow shop scheduling problem,” Journal of
King Saud University - Computer and Information Sciences, vol. 34, no. 9, 2022,
doi: 10.1016/j.jksuci.2021.08.025.

[95] K. Drachal and M. Pawłowski, “A review of the applications of genetic algorithms

to forecasting prices of commodities,” Economies, vol. 9, no. 1, 2021, doi:
10.3390/economies9010006.

[96] H. Zhi and S. Liu, “Face recognition based on genetic algorithm,” J Vis Commun
Image Represent, vol. 58, 2019, doi: 10.1016/j.jvcir.2018.12.012.

[97] P. Yarsky, “Using a genetic algorithm to fit parameters of a COVID-19 SEIR

model for US states,” Math Comput Simul, vol. 185, 2021, doi:
10.1016/j.matcom.2021.01.022.

[98] M. Mitchell, An introduction to genetic algorithms., vol. 8. in Complex adaptive

systems., vol. 8. Cambridge, MA, US: The MIT Press, 1996.

[99] T. Alam, S. Qamar, A. Dixit, and M. Benaida, “Genetic algorithm: Reviews,

implementations and applications,” International Journal of Engineering
Pedagogy, vol. 10, no. 6. 2021. doi: 10.3991/IJEP.V10I6.14567.

[100] “A STUDY ON GENETIC ALGORITHM AND ITS APPLICATIONS,”

International Research Journal of Modernization in Engineering Technology and
Science, 2023, doi: 10.56726/irjmets32980.

[101] K. F. Man, K. S. Tang, and S. Kwong, “Genetic algorithms: concepts and

applications [in engineering design],” IEEE Transactions on Industrial
Electronics, vol. 43, no. 5, pp. 519–534, 1996, doi: 10.1109/41.538609.
 142

[102] R. Ingo, “Evolution strategy: Optimization of technical systems by means of

biological evolution. Fromman-Holzboog,” Stuttgart, vol. 104, p. 15, 1973.

[103] H. Fuger, G. Stein, and U. Stilla, “Multi-population evolution strategies for

structural image analysis,” in Proceedings of the First IEEE Conference on
Evolutionary Computation. IEEE World Congress on Computational Intelligence,
1994, pp. 229–234 vol.1. doi: 10.1109/ICEC.1994.350011.

[104] J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proceedings of

ICNN’95 - International Conference on Neural Networks, IEEE, pp. 1942–1948.
doi: 10.1109/ICNN.1995.488968.

[105] Q. Bai, “Analysis of Particle Swarm Optimization Algorithm,” Comput. Inf. Sci.,
vol. 3, pp. 180–184, 2010.

[106] H. Ahmed and J. Glasgow, “Swarm intelligence: concepts, models and

applications,” School Of Computing, Queens University Technical Report, 2012.

[107] T. M. Mmola, A. S. Nhleko, and J. M. Atherfold, “Application of Particle Swarm

Optimization Algorithm to Optimize Stope Layout for Underground Mines,” in
Proceedings of the 27th International Symposium on Mine Planning and
Equipment Selection - MPES 2018, Springer International Publishing, 2019, pp.
213–221. doi: 10.1007/978-3-319-99220-4_17.

[108] R. C. Eberhart and Y. Shi, “Guest Editorial Special Issue on Particle Swarm
Optimization,” IEEE Transactions on Evolutionary Computation, vol. 8, no. 3, pp.
201–203, 2004, doi: 10.1109/TEVC.2004.830335.

[109] E. S. Olivas, J. D. M. Guerrero, M. Martinez-Sober, J. R. Magdalena-Benedito,

and L. Serrano, Handbook of research on machine learning applications and
trends: Algorithms, methods, and techniques: Algorithms, methods, and
techniques. IGI global, 2009.

[110] C. P. Lim and S. Dehuri, Innovations in swarm intelligence, 2010th ed., vol. 248.
Springer Science & Business Media, 2009.
 143

[111] K. E. Parsopoulos and M. N. Vrahatis, “Particle swarm optimization method for

constrained optimization problems,” Intelligent technologies–theory and
application: New trends in intelligent technologies , vol. 76, no. 1, pp. 214–220,
2002.

[112] I. Rahman, P. Vasant, B. Singh, and M. Abdullah-Al-Wadud, “On the performance

of accelerated particle swarm optimization for charging plug-in hybrid electric
vehicles,” Alexandria Engineering Journal, vol. 55, pp. 419–426, Mar. 2016, doi:
10.1016/j.aej.2015.11.002.

[113] V. R. Kulkarni and V. Desai, “ABC and PSO: A comparative analysis,” in 2016
IEEE International Conference on Computational Intelligence and Computing
Research (ICCIC), IEEE, 2016, pp. 1–7.

[114] R. J. Vanderbei, “LINEAR PROGRAMMING: Foundations and Extensions,”

Journal of Operations Research Society, vol. 49, p. 94, 1998.

[115] H. P. Williams, Model building in mathematical programming, 5th ed. John Wiley
& Sons, 2013. Accessed: Jun. 11, 2023. [Online]. Available:
https://books.google.com/books?id=0tR8tcwYCM4C

[116] G. B. Dantzig and M. N. Thapa, Linear programming: Theory and extensions,

2003rd ed., vol. 2. Springer, 2003.

[117] E. K. Chong and H. Ż. Stanislaw, AN INTRODUCTION TO OPTIMIZATION, vol.

75. John Wiley & Sons, 2013.

[118] D. Luenberger and Y. Ye, Linear and Nonlinear Programming. 2021. doi:
10.1007/978-3-030-85450-8.

[119] A. Alochukwu et al., “stope boundary optimisation for underground mines,”

Group (New York), 2018.

[120] Gurobi Optimization LLC, “Gurobi Optimizer Reference Manual.” 2023. [Online].
Available: https://www.gurobi.com

[121] E. L. Johnson, Modeling and strong linear programs for mixed integer
programming. Springer, 1989.
 144

[122] L. Wolsey, “Strong formulations for mixed integer programming: a survey,” Math
Program, vol. 45, pp. 173–191, 1989.

[123] M. Asghari, A. M. Fathollahi-Fard, S. M. J. Mirzapour Al-e-hashem, and M. A.

Dulebenets, “Transformation and Linearization Techniques in Optimization: A
State-of-the-Art Survey,” Mathematics, vol. 10, no. 2, 2022, doi:
10.3390/math10020283.

[124] W. P. Adams and H. D. Sherali, “A tight linearization and an algorithm for zero-
one quadratic programming problems,” Manage Sci, vol. 32, no. 10, pp. 1274–
1290, 1986.

[125] X. He, A. Chen, W. A. Chaovalitwongse, and H. X. Liu, “An improved

linearization technique for a class of quadratic 0-1 programming problems,” Optim
Lett, vol. 6, no. 1, 2012, doi: 10.1007/s11590-010-0249-z.

[126] H. D. Sherali and A. Alameddine, “A new reformulation-linearization technique

for bilinear programming problems,” Journal of Global Optimization, vol. 2, no. 4,
1992, doi: 10.1007/BF00122429.

[127] B. Rostami and F. Malucelli, “A revised reformulation-linearization technique for

the quadratic assignment problem,” Discrete Optimization, vol. 14, 2014, doi:
10.1016/j.disopt.2014.08.003.

[128] A. Selvi, D. den Hertog, and W. Wiesemann, “A reformulation-linearization

technique for optimization over simplices,” Math Program, vol. 197, no. 1, 2023,
doi: 10.1007/s10107-021-01726-y.

[129] C. Dapogny, A. Faure, G. Michailidis, G. Allaire, A. Couvelas, and R. Estevez,

“Geometric constraints for shape and topology optimization in architectural
design,” Comput Mech, vol. 59, no. 6, 2017, doi: 10.1007/s00466-017-1383-6.

[130] M. Dagbert, “Review of the Book ‘Mineral Resource Estimation’ by Mario Rossi
and Clayton Deutsch, 332p, Published by Springer in Early 2014 (ISBN 978-1-
4020-5716-8),” Math Geosci, vol. 46, no. 8, pp. 1029–1034, 2014, doi:
10.1007/s11004-014-9547-9.
 145

[131] J. Poniewierski, “BLOCK MODEL KNOWLEDGE FOR MINING ENGINEERS-

AN INTRODUCTION.” [Online]. Available:
www.linkedin.com/in/julianponiewierski/

[132] M. Bustillo Revuelta, “Mineral Resource Evaluation,” in Mineral Resources:

From Exploration to Sustainability Assessment, M. Bustillo Revuelta, Ed., Cham:
Springer International Publishing, 2018, pp. 223–309. doi: 10.1007/978-3-319-
58760-8_4.

[133] R. L. Lowrie, “SME Mining Reference Handbook,” Society for Mining,

Metallurgy, and Exploration (SME). Society for Mining, Metallurgy, and
Exploration (SME), 2002. [Online]. Available:
https://app.knovel.com/hotlink/toc/id:kpSMEMRH03/sme-mining-reference/sme-
mining-reference

[134] V. Rudenno, The mining valuation handbook mining and energy valuation for
investors and management. 4th ed. Milton, Qld. : John Wiley, 2012., 2012.
[Online]. Available: https://search.library.wisc.edu/catalog/9911077337702121

[135] A. Esmaeili, J. K. Hamidi, and A. Mousavi, “Determination of sublevel stoping

layout using a network flow algorithm and the MRMR classification system,”
Resources Policy, vol. 80, p. 103265, Jan. 2023, doi:
10.1016/J.RESOURPOL.2022.103265.

[136] L. Sipeki, A. M. Newman, and C. A. Yano, “Selecting support pillars in

underground mines with ore veins,” IISE Trans, vol. 52, no. 10, pp. 1173–1188,
Oct. 2020, doi: 10.1080/24725854.2019.1699978.

[137] The MathWorks, “MATLAB version: 9.13.0 (R2022b).” The MathWorks Inc.,
Natick, Massachusetts, United States, 2022. [Online]. Available:
https://www.mathworks.com

[138] MathWorks, “Problem-Based Optimization Workflow.” [Online]. Available:

https://www.mathworks.com/help/optim/ug/problem-based-workflow.html

[139] B. King and A. Newman, “Optimizing the cutoff grade for an operational
underground mine,” Interfaces (Providence), vol. 48, no. 4, pp. 357–371, 2018.
 146

[140] F. Sotoudeh, M. Nehring, M. Kizil, P. Knights, and A. Mousavi, “A novel cut-off

grade method for increasing the sustainability of underground metalliferous
mining operations,” Miner Eng, vol. 172, p. 107168, 2021, doi:
https://doi.org/10.1016/j.mineng.2021.107168.

[141] R. Martinelli, J. Collard, and M. Gamache, “Strategic planning of an underground

mine with variable cut-off grades,” Optimization and Engineering, vol. 21, no. 3,
pp. 803–849, 2020, doi: 10.1007/s11081-019-09479-6.
 147

VITA

Theophilus Mensah was born in Ghana, West Africa, and received his Bachelor of

Science in Mining Engineering from the University of Mines and Technology “UMaT”,

Tarkwa, Ghana, in June 2016. He moved on to work with Perseus Mining Ghana Limited

as a Mining Engineer until July 2021. In July 2021, he moved to the USA where he earned

his Master of Science in Mining and Explosives Engineering from the Missouri University

of Science and Technology “MST”, Rolla in December 2023. Prior to completion, he

accepted an offer to work for Freeport McMoran as a Mine Planning Engineer.