International Journal of Mining Science and Technology 25 (2015) 767–772

Contents lists available at ScienceDirect

International Journal of Mining Science and Technology

journal homepage: www.elsevier.com/locate/ijmst

Designing an optimal stope layout for underground mining based on a

heuristic algorithm
Sandanayake Don Suneth Sameera ⇑, Topal Erkan, Asad Mohammad Waqar Ali
Mining Engineering, Western Australian School of Mines, Curtin University, Kalgoorlie, WA 6433, Australia

a r t i c l e i n f o a b s t r a c t

Article history: An optimal layout or three-dimensional spatial distribution of stopes guarantees the maximum prof-
Received 9 December 2014 itability over life span of an underground mining operation. Thus, stope optimization is one of the key
Received in revised form 12 April 2015 areas in underground mine planning practice. However, the computational complexity in developing
Accepted 20 May 2015
an optimal stope layout has been a reason for limited availability of the algorithms offering solution to
Available online 28 August 2015
this problem. This article shares a new and efficient heuristic algorithm that considers a
three-dimensional ore body model as an input, maximizes the economic value, and satisfies the physical
Keywords:
mining and geotechnical constraints for generating an optimal stope layout. An implementation at a cop-
Stope optimization
Underground mining
per deposit demonstrates the applicability and robustness of the algorithm. A parallel processing based
Underground mine planning modification improving the performance of the original algorithm in terms of enormous computational
Heuristic algorithm time saving is also presented.
Ó 2015 Published by Elsevier B.V. on behalf of China University of Mining & Technology.

1. Introduction and as a result, generating solution to this problem, is a challenge.

The availability of adequate supply of mineral resources is the
requirement for commencement of a mining operation. Geological
and geostatistical investigations delineate the spatial (size, shape,
depth, etc.) characteristics of these resources, define mineable
reserves, and create a three-dimensional ore body model by divid-
ing these reserves into thousands of mining blocks [1–5]. Apart
from the economic parameters, this ore body model becomes an
input to subsequent mine planning process. Given these inputs,
mine planning process may suggest recovery of these reserves
either through a surface or an underground mining operation. If
the decision is in favor of an underground mining operation, then
the development of an optimal production plan that maximizes
the discounted value of the operation subject to the production
capacity, physical, and geotechnical constraints, follows the selec-
tion of a stope layout [6,7]. The procedure to generate an optimal
stope layout combines thousands of mining blocks into a set of
stopes, such that, the undiscounted value of operation is maxi-
mized by satisfying the physical and geotechnical requirements
[8,9]. Consequently, for a decision, whether a mining block is
included in a stope or otherwise, it requires evaluation of all possi-
ble combinations of thousands of mining blocks. This establishes
the computational complexity of the stope optimization problem,
⇑ Corresponding author. Tel.: +61 890886150.
E-mail address: Donsuneth.Sandanayake@curtin.edu.au (D.S.S. Sandanayake).
Therefore, this paper describes and implements an efficient heuris-
tic approach to develop an optimal stope layout in underground
mining operations.
Physical and geotechnical constraints relate to the size and ori-
entation of the ore body, extent of mine development openings
(levels) that provide access to stopes, mining equipment size, and
an appropriate ore pillar size ensuring stability of these under-
ground excavations [10,11]. Given these constraints, a realistic
stope layout may constitute fixed (i.e. all stopes of similar size)
or variable size stopes with or without pillars. Consequently, a pro-
cedure that caters for variable stope sizes and availability of pillars
within the underground mine, supplements the computational
complexity of the stope layout problem. Fig. 1 establishes the con-
text of computational complexity in a simple two-dimensional
hypothetical ore body model containing 64 mining blocks.
As shown in Fig. 1a, the stope size is defined as 3  3 = 9 mining
blocks, i.e. the number of mining blocks along x and y-axis is equal
to 3, respectively. Given the defined stope size, a candidate mining
block b may become part of 1 out of 9 possible stope combinations
(for example, stopes 1 and 2). This reflects that there are numerous
possible combinations for all 64 mining blocks within this hypo-
thetical ore body model. Similarly, Fig. 2b relates the stope size
and possible stope combinations for a mining block b. It shows that
an increase in stope size escalates the number of stope combina-
tions, resulting in the computational complexity of the stope lay-
out problem. In realistic ore body models, an evaluation of these

http://dx.doi.org/10.1016/j.ijmst.2015.07.011
2095-2686/Ó 2015 Published by Elsevier B.V. on behalf of China University of Mining & Technology.
768 D.S.S. Sandanayake et al. / International Journal of Mining Science and Technology 25 (2015) 767–772
Fig. 1. Complexity of stope layout problem.
Fig. 2. An illustration of the stope and pillar layout notations.
combinations is required in three-dimensional space, leading to an
exponential increase in the computational time and complexity.
Moreover, if a mining block is shared among a number of pos-
sible stopes, i.e. the mining block exists in more than one stope
sets, such combinations are categorized as overlapping stopes.
Physical mining constraint restricts the generation of overlapping
stopes, and accordingly, avoiding such stope sets requires imple-
mentation of the additional computational steps of the algorith-
mic. In summary, it is established that the computational
complexity of the stope layout problem is manifolds, and the
development of a stope optimization algorithm that evaluates mul-
tiple combinations of non-overlapping stope sets and selects the
best stope layout is a challenge.
Given these intricacies, a few algorithms offer solution to the
stope layout problem, however, a majority of these algorithms
does not generate the optimum solution [12]. Ovanic and Young
propose the branch and bound algorithm to optimize the stope
boundary along a row of mining blocks, i.e., in one dimension,
using two piecewise linear cumulative functions [13,14]. These
two functions identify the optimal starting and ending locations
for mining within each row of mining blocks. As such, the algo-
rithm cannot be implemented in three-dimensional ore body mod-
els. Imitating the open pit optimization procedures, Alford shares a
floating stope procedure, and given the structure of the algorithm,
it violates the non-overlapping stopes requirement [15]. Cawrse
suggests a multi-pass floating stope process as an improve-
ment/extension in the original floating stope process; however,
the algorithm fails to address the violation of non-overlapping
stopes requirement [16]. Ataee Pour proposes the maximum
neighborhood value algorithm that relies on individual mining
blocks in delineating the stope boundaries, consequently, ignores
the shape of the mineable stopes, leading to limited applicability
in realistic scenarios [17]. Grieco and Dimitrakopoulos develop a
mixed integer programming based stope optimization algorithm
under geological uncertainty, and given the size of the stochastic
framework, the algorithm focuses only on the profitable stope
locations, and valid solution to the entire ore body does not exist
[18]. Topal and Sens suggest a heuristic procedure that derives
the most profitable stopes from the ore body model, however,
the procedure fails to analyse all alternative solutions [9]. Bai
et al. propose an implementation of the maximum flow algorithm
for stope optimization problem [19]. However, it is limited to use
for small mineralized ore bodies and sub-level stoping mining
method.
Realizing that the earlier studies do not offer a holistic approach
to solve this challenging problem, this paper contributes: (1) a new
and efficient heuristic algorithm that maximizes the economic
value of the operation, honors the physical mining and geotechni-
cal constraints, incorporates fixed and variable stope sizes with
and without pillars, and solves the problem in three-dimensional
space; (2) an implementation of the original algorithm at a copper
deposit; and (3) a parallel processing based modified algorithm
improving the performance of the original procedure in terms of
enormous computational time savings.
2. Proposed heuristic algorithm
The proposed heuristic algorithm solves the stope layout prob-
lem sequentially in five distinct steps. It standardizes the ore body
model, creates stopes, assigns attributes to these stopes, generates
sets of stopes as possible solutions, and identifies the optimal solu-
tion among all possible solutions. More specifically, it converts the
ore body model into an economic block model that constitutes
 D.S.S. Sandanayake et al. / International Journal of Mining Science and Technology 25 (2015) 767–772 769

Fig. 3. Graphical representation of variable stope sizes in algorithmic step 2.

mb = mining block density, ton/m3;

eb = mining block economic value in dollars;
gb = mining block grade, % or g/ton;
p = pillar width, meters;
xs, ys, zs = x, y, and z coordinates of stope s;
Ls, Ws, Hs = length, width, and height (meters) of stope s;
ms = material density of stope s, ton/m3;
es = economic value of stope s, dollars;
gs = grade of stope s, % or g/ton;
cxyzs ¼ x; y; z coordinates of the origin mining block in stope s;
lxyzs ¼ x; y; z coordinates of the terminal (last) mining block in
stope s;
ef = economic value of the stope set f, dollars;

Fig. 2 illustrates some of these indices, parameters and

variables.
Given these notations, the algorithm is implemented as follows:

(1) Define mining block length, width and height, i.e., Lb, Wb, Hb,
select the minimum and maximum coordinates of the given
irregular block model, i.e., min. (xb, yb, zb) and max. (xb, yb,
zb), and translate the irregular block model into a regularised
model with these parameters.
(2) Convert the regularised ore body model into an economic
block model, i.e. calculate the block economic value
Fig. 4. Stages 4 and 5 of the stope optimiser algorithm. eb ¼ ½ðP  r b Þ  g b  y  ðM b þ cb Þ  tb .
(3) Define oxs, oys, ozs and ix, iy, iz to create a set ds of stopes with
es > 0 and variable stope sizes (vxs, vys, vzs). Each stope s e ds
mining blocks with consistent sizes, combines mining blocks into constitutes a set of mining blocks bs. Here, vxs = (oxs + i0 x),
stopes based on defined stope sizes (fixed or variable), assigns 0
vys = (oys + i0 y) and vzs = (ozs + i0 z), such that 0 6 i x 6 ix,
attributes such as grade and material density to these stopes, gen- 0 0
0 6 i y 6 iy and 0 6 i z 6 iz.
erates sets of positive-value stopes as all possible combinations of
solutions, and filters these sets of stopes through mining con-
For a two-dimensional example, if oxs, oys, ix, and iy are set at 2,
straints such as availability of development openings (levels) and 0 0
2, 2, and 0, respectively, such that, 0 6 i x 6 2 and 0 6 i y 6 0. Thus,
pillars for identifying the optimal solution. 0 0
Following notations facilitate the description of the algorithm. i x ¼ f0; 1; 2g and i y ¼ f0g. Consequently, variable stope sizes
0 0
include, v xs ; v ys ¼ ðoxs þ i xÞ; ðoys þ i yÞ ¼ ð2 þ 0Þ; ð2 þ 0Þ ¼ ð2; 2Þ,
0 0
b = mining block indicator; v xs ; v ys ¼ ðoxs þ i xÞ; ðoys þ i yÞ ¼ ð2 þ 1Þ; ð2 þ 0Þ ¼ ð3; 2Þ, and
s = stope indicator; v xs ; v ys ¼ ðoxs þ i0 xÞ; ðoys þ i0 yÞ ¼ ð2 þ 2Þ; ð2 þ 0Þ ¼ ð4; 2Þ. Fig. 3
f = stope set indicator; demonstrates a graphical representation of this concept.
k = stope level indicator; P
P = metal price ($/ton) mb P
(4) Calculate stope attributes ms ¼ b
jbs j
; es ¼ b eb , and
Mb = mining cost ($/ton) P
ðmb g b Lb W b Hb Þ
cb = processing cost ($/ton) g s ¼ Pb ðm L W H Þ .
b b b b b
rb = refining cost ($/ton)
(5) Create a family of subsets F f over ds as given below:
y = recovery (%)
(a) Evaluate the mining constraint
vxs, vys, vzs = variable stope size in terms of number mining  
cxyzs P lxyzs0 \ cxyzs0 P lxyzs ; 8s; s0 2 ds , generate #f as sets of
blocks along x, y, and z-axis;
oxs, oys, ozs = minimum stope size in terms of number of mining non-overlapping stopes if pillar width and level height are
blocks along x, y, and z-axis; not defined, and update F f ¼ fF f [ #f g.

ix, iy, iz = maximum incremental number of mining blocks along (b) Evaluate the mining constraint jcxs  cxs0 j [ jcys  cys0 j P
x, y, and z-axis for stope size variation; p; 8s; s0 2 dsk , generate #f sets of stopes (with pillar width
xb, yb, zb = mining block x, y, and z coordinates; apart) if a pillar width and level height are defined, and
Lb, Wb, Hb = mining block length, width, and height in meters; update F f ¼ fF f [ #f g:
770 D.S.S. Sandanayake et al. / International Journal of Mining Science and Technology 25 (2015) 767–772

Table 1 Case C: variable stope size without pillar definition, keeping

Stope attributes in Cases A, B, C and D. oxs = 3, oys = 3, ozs = 3, and ix = 1, iy = 0, iz = 0, stope sizes
Case | ds | ms es gs 3  3  3 and 4  3  3 without pillars between stopes.
Min Max Min Max Min Max Case D: variable stope size with pillar definition, keeping oxs = 3,
oys = 3, ozs = 3, and ix = 1, iy = 0, iz = 0, i.e. stope sizes 3  3  3
A 4747 2.62 3.86 2575,844 27141496 0 4.19
B 1530 2.62 3.83 2575,844 26660108 0 4.06
and 4  3  3 with pillars between stopes.
C 9215 2.62 3.87 3434,459 35277396 0 4.19
D 3090 2.62 3.83 3434,459 35277396 0 4.07 For each defined case, the algorithm implements steps 4, 5, and
6, i.e. it combines mining blocks into a set of all possible stopes,
generates a set of positive-valued stopes, develops feasible solu-
tions by filtering these stopes through physical mining constraints,
and finally among all feasible solutions, it solves for the optimal
solution, i.e. a stope layout with the maximum economic value.
Table 1 presents the attributes of the generated stopes in Cases
A, B, C, and D, and the extracted number of positive-valued stopes
in these cases.
The initial step in the optimization process, i.e. application of
mining constraints to positive-valued blocks, generates infinite
sets of stopes. Thus, achieving an optimal solution in all four cases
becomes impossible. Consequently, we define an upper bound (nf)
on the number of possible solutions, and execute the algorithm
Fig. 5. Convergence of the solution value for Cases A, B, C and D. over a number of iterations until convergence of the solution value.
Each iteration increments nf, and maintains the highest-valued
P solution, performing a comparison of new and previous solutions.
(6) Determine ef ¼ s es for each non-overlapping stope set
Fig. 5 presents the increase and convergence in the solution values
#f 2 F f , and select the optimal set #f 0 2 F f , such that
for Cases A, B, C and D. Also, Table 2 presents the attributes of the
ef 0 ¼ maxðef Þ. optimum solutions in Cases A, B, C and D.
Fig. 6 shows the three-dimensional view of the optimal stope
Fig. 4 demonstrates the stage 4 and 5 of the stope optimizer layouts that were visualized using the Gemcom Surpac™ mine
algorithm through a flow chart. design software version 6.3.1 (GEOVIA Surpac™ 2012) for Cases
A, B, C and D, respectively.
3. Implementation of proposed algorithm at a copper deposit

This section describes an implementation of the algorithm in an

ore body model for a copper deposit. The ore body model consists
of 47,052 mining blocks of irregular size (both cubic and cuboid
shape), ranging from 20 m  20 m  0.2 m to 20 m  20 m  20 m.
The minimum, average, and maximum grade of ore is 0%, 0.021%,
15.32%, respectively. Similarly, the material density ranges from
2.62 to 4.66 ton/m3. The standardized ore body model constitutes
287,984 regular sized 10 m  10 m  10 m mining blocks, with
minimum, average, and maximum grade of ore is 0%, 0.021%,
8.91%, respectively, and the material density ranging from 2.62 to
4.51 ton/m3. This regularized ore body model is then converted
into an economic block model. For this conversion, the copper
price, mining cost and recovery were assumed to be 8000 $/ton,
30 $/ton and 100% respectively, with negligible refining and pro-
cessing costs.
Given the economic block model, we present the performance
of the algorithm in following four realistic cases:
Case A: fixed stope size without pillar definition, keeping oxs = 3,
oys = 3, ozs = 3, i.e. a stope size of 3  3  3 without pillars
between stopes.
Case B: fixed stope size with pillar definition, keeping oxs = 3,
oys = 3, ozs = 3, i.e. a stope size 3  3  3 with pillars between
stopes.
3.1. Modified algorithm with parallel processing
A performance evaluation of the algorithm in Cases A, B, C, and
D reveals that an increase in the upper bound of the set count or
possible solutions results in an exponential increase in solution
time. Fig. 7 confirms this relationship between nf and the solution
time for Cases A, B, C and D. Consequently, we resolve this algorith-
mic complexity using the task parallelization concept in
object-oriented programming, and propose a change in the struc-
ture of the original stope optimization algorithm. This requires a
division of the set F f into two unique subsets l1f ; l2f  F f , such
that l1f \ l2f ¼ føg and l1f ] l2f ¼ F f . Fig. 8 illustrates the struc-
ture of the proposed modification in the original algorithm.
As shown in Fig. 9, the modified algorithm performs a parallel
implementation of step 4 on subsets l1f and l2f , requiring less
computations as compared to the execution of the original algo-
rithm on set F f . Consequently, this asynchronous execution of
the algorithm helps achieve the required performance gain. Simi-
larly, the modified algorithm improves the performances in Cases
B, C and D. Table 3 demonstrates a comparison of the optimal solu-
tion attributes in modified and original algorithms implemented in
Cases A, B, C and D. Similarly, Fig. 9 presents a performance profile
of the modified and original algorithms implemented in Case C.

Table 2
Comparison of solution parameters for Cases A, B, C and D using the proposed algorithm.

Parameter Solution value ($) Average stope value ($) Average grade (%) Average ton (ton) No. of stopes Solution time (h: min: s)
Case A 779,872,119 4,456,412 1.0120 87438 175 07:00:00
Case B 436,341,939 5,073,743 1.1010 42904 86 00:00:50
Case C 773,069,150 5,368,535 0.9895 89856 144 00:01:10
Case D 382,129,243 5,619,547 1.0146 42668 68 00:06:05
 D.S.S. Sandanayake et al. / International Journal of Mining Science and Technology 25 (2015) 767–772 771

5875 E

5900 E
6000 E
5975 E

6100 E
4800 N 4800 N

6075 E
4900 N 4900 N
6175 E

125 Z
5000 N 5000 N
125 Z
5100 N 5100 N
5200 N 5200 N

25 Z 25 Z
4800 N

4800 N
4900 N

4900 N
75 Z 75 Z
5000 N

5000 N
es ($)
5100 N

5100 N
5×10 6
5200 N

5200 N
5-10×10 6
5-15×10 6
15×10 6
(a) Case A (b) Case B
5875 E

5900 E
6000 E
5975 E

6100 E
4800 N 4800 N
6075 E

4900 N 4900 N
6175 E

125 Z
5000 N 5000 N
125 Z
5100 N 5100 N
5200 N 5200 N

25 Z 25 Z
4800 N

4800 N
4900 N

4900 N
75 Z 75 Z
5000 N

5000 N
5100 N

5100 N
5200 N

5200 N
(c) Case C (d) Case D

Fig. 6. Three-dimensional view of the optimal stope layouts in Cases A, B, C, and D.

2000 2000

Case A Modified-Case C
Solution time (min)

1600 1600
Solution time (min)

Case B Original-Case C
1200 Case C 1200
Case D
800 800

400 400
0
0
1 3 5 7 9 11 13 15 17 19
500 1500 2500 3500 4500 5500 6500 7500 8500 9500
nf ($)
nf ($)
Fig. 7. Relationship between the set count and solution time for Cases A, B,C and D.
Fig. 9. Performance gain in Case C by the modified algorithm.

Table 3 shows that Case A generates the best solution among

four cases. Although, the stope size is identical in Cases A and B,
the solution value in Case B is significantly less than in Case A.
Defining pillars and levels prior to the implementation of the opti-
mization step results in a reduction of the number of stopes in the
solution. As such, the number of stopes is reduced from 175 in Case
A to 86 in Case B, and consequent a decrease in the solution value.
Similarly, an option to include of pillars and levels becomes a rea-
son for the solution values compared in Cases C and D. However,
Cases B, C, and D converge at nf = 500, 500, and 1000 respectively,
requiring less iteration to achieve optimal solutions to these cases.
Case A converges at nf = 7000 and consequently, it requires a
higher number of iterations than for Cases B, C and D in order to
achieve the optimal solution. As such, in comparison of solution
times among cases, the solution time for Case A is significantly
Fig. 8. Structure of the parallelized processes in modified algorithm. higher as opposed to Cases B, C and D.
772 D.S.S. Sandanayake et al. / International Journal of Mining Science and Technology 25 (2015) 767–772
Table 3
Comparison of solution parameters for Cases A, B, C and D using the modified algorithm.
Parameter Solution value ($) Average stope value ($) Average grade (%)
Case A 779,872,119 4,456,412 1.0120
Case B 436,341,939 5,073,743 1.1010
Case C 773,069,150 5,368,535 0.9895
Case D 382,129,243 5,619,547 1.0146
4. Conclusions
This article shares a heuristic algorithm for the development of
an optimal stope layout in an underground mining operation. The
proposed algorithm incorporates variable stope sizes with or with-
out pillars, maximizes the value of the operation, and satisfies the
mining and geotechnical constraints. Unlike previous studies, the
algorithm offers a solution in three-dimensional space, generates
practical non-overlapping stopes, and it is flexible enough for
applications in varying underground mining situations.
An implementation at a copper deposit in four different cases
demonstrates the robustness of the algorithm, and based on the
performance evaluation in these cases, a parallel processing based
modification in the original algorithm improves the efficiency in
terms of significantly reduced solution times. Thus, the proposed
heuristic method addresses the established computational com-
plexity of the stope optimization problem. However, unlike the
conventional xyz orientation of the mining blocks as input, an eval-
uation of other possible orientations and their impact on the value
of the optimal stope layout provides a direction for future research.
References
[1] Asad MWA, Dimitrakopoulos R, Eldert JV. Stochastic production phase design
for an open pit mining complex with multiple processing streams. Eng
Optimization 2014;46(8):1139–52.
[2] Hustruilid W, Kuchta M. Open pit mine planning & design. Netherlands
(Leiden): Taylor and Francis; 2006.
[3] Kumral M. Production planning of mines: optimisation of block sequencing
and destination. Int J Mining Reclamat Environ 2006;26:93–103.
[4] Little J, Topal E, Knights P. Simultaneous optimisation of stope layouts and long
term production schedules. Mining Technol 2011;120:129–36.
Average tonnage (ton) No. of stopes Solution time (h: min: s)
87438 175 03:21:00
42904 86 00:00:20
89856 144 00:00:30
42668 68 00:02:45
[5] Leite A, Dimitrakopoulos R. Stochastic optimization of mine production
scheduling with uncertain ore/metal/waste supply. Int J Mining Sci Technol
2014;24:755–62.
[6] Kumral M. Multi-period mine planning with multi-process routes. Int J Mining
Sci Technol 2013;23:317–21.
[7] Opoku S, Cuthbert M. Stochastic modelling of the open pit to underground
transition interface for gold mines. Int J Mining Reclamat Environ
2013;27:407–24.
[8] Little J, Topal E. Strategies to assist in obtaining an optimal solution for an
underground mine planning problem using mixed integer programming. Int J
Mining Mining Eng 2011;3:152–72.
[9] Topal E, Sens J. A new algorithm for stope boundary optimisation. J Coal Sci Eng
2010;16:113–9.
[10] Alford C, Brazil M, Lee D. Handbook of operations research in natural
resources. Germany: Springer; 2007. p. 561–78.
[11] Gertch RE, Bullock RL. General mine planning in techniques in underground
mining. Soc Mining Metall Explor USA 1998:87–127.
[12] Ataee Pour M. A critical survey of the existing stope layout optimization
techniques. J Mining Sci 2005;41:447–66.
[13] Ovanic J, Young DS. Economic optimization of stope geometry using separable
programming with special branch and bound technique. In: Proceedings of 3rd
Canadian conference on computer applications in the minerals industry.
Quebec; 1995. p. 129–35.
[14] Ovanic J, Young DS. Economic optimization of open stope geometry. In:
Proceeding of 28th International APCOM Symposium. Colorado school of
Mines, Golden, Colorado; 1999. p. 855–62.
[15] Alford C. Optimisation in underground mine design. In: Proceeding of 25th
International APCOM Symposium. Brisbane; 1995. p. 213–8.
[16] Cawrse I. Multiple pass oating stope process. In: Proceedings of Strategic Mine
Planning Conference. Perth; 2001. p. 87–93.
[17] Ataee-Pour M. A heuristic algorithm to optimise stope boundaries. New South
Wales (Australia): University of Wollongong; 2000. p. 3–36.
[18] Grieco N, Dimitrakopoulos R. Managing grade risk in stope design
optimisation: probabilistic mathematical programming modal and
application in sublevel stoping. Mining Technol 2007;116:49–57.
[19] Bai X, Marcotte D, Simon R. Underground stope optimization with network ow
method. Comput Geosci 2012;52:361–71.