Strategic mine planning and design: some

challenges and strategies for addressing them

P. A. Dowd1*, C. Xu1 and S. Coward2

The authors provide their assessment of current challenges in strategic mine planning and some
approaches for addressing them. Challenges covered are quantification of downstream processes and
their integration into orebody models; spatial characterisation of geometallurgical variables and their
integration into block models, scheduling and mine optimisation; spatial characterisation of new variables
for new types of mining; flexibility in planning and design to manage risk and minimise its impact; IT
infrastructure for rapid on-line data collection and processing. Most of these challenges require new
types of data, variables, modelling and estimation methods. Foremost among new variable types
are geometallurgical and dynamic rock mass characterisation variables. New types of data and data
collection include rapid generation of very large amounts of on-line sensor data and the consequent
need for rapid processing and modelling of these data. The paper includes examples of models and
outputs to illustrate approaches and potential solutions.
Keywords: Geometallurgy, Mine optimisation, Block caving, Solution mining, Quantification of downstream processes, Fracture modelling
The paper is part of a special issue on Smart Mining Complexes and Mineral Value Chains

Introduction propagation model and, for solution mining, flow modelling

through engineered fracture systems. These models are also
In this paper, we address five specific areas that we have cho- relevant to the stability of excavations.
sen to call challenges and we outline largely our own attempts Challenge (4) requires an integrated orebody/extraction
to address them. These challenges are: model that maximises flexibility in design across a range of
(1) Realistic quantification of downstream processes applied to alternative optima for changes in prices, costs and technical
orebody models to provide an integrated approach to mine
variables.
design and optimisation.
(2) Modelling, estimation and simulation of geometallurgical While the infrastructure requirement in Challenge (5)
variables and their integration into resource and reserve esti- is an IT and industry/government investment problem, the
mation and mine planning. adaptation of current resource and reserve estimation and
(3) Spatial modelling, estimation and simulation of rock charac- optimal production methods to massive volumes of continu-
teristics for new types of mining – deep mining, particularly ously sensed proxy data is highly relevant to strategic mine
block caving, and solution mining. planning.
(4) Flexibility in planning and design to manage risk and min- Two of the most significant issues for mining operations
imise its impact. are cost reduction (especially energy reduction) and geome-
(5) IT infrastructure and platforms for rapid on-line data col- tallurgy (much of which is related to energy reduction). An
lection, storage, access and processing.These challenges are
example of both is optimal fragmentation of rock in blasting
not all independent and a solution to any one of them may
require the simultaneous solution of part or all of another. to minimise down-stream rock breakage costs, optimise pack-
Challenge (1) requires modelling specific processes (e.g., ing in trucks, optimise ore selection and reduce processing
selection, loading, blending, physical and chemical process- costs. Processing and breakage require geometallurgical var-
ing), integrating them with orebody models and estimating the iables and modelling.
impacts of the processes at each stage of the mining operation. Another major area is the increasing focus on deep min-
Challenge (2) covers a wide range of new variables to be ing and, in particular, large-scale block and panel caving.
included in resource and reserve modelling and estimation and The deep mining focus is accompanied by developments in
requires integration with downstream processes as in Challenge deep exploration, such as those being led by the Australian
(1). Many of these variables are non-additive and, in some cases, Co-operative Research Centre (CRC) for Deep Drilling
are measured indirectly by sensing and/or by proxy variables. Technologies. Both deep exploration and deep mining require
The fundamental requirement in Challenge (3) is a sto- new approaches to resource/reserve modelling, estimation and
chastic model of in situ fractures together with a fracture simulation and to process modelling.
Geometallurgy brings new challenges: non-additive varia-
1
The University of Adelaide, Adelaide, SA, Australia
bles often with very little data; difficulties in up-scaling from
2
QG Consulting, Perth, WA, Australia the laboratory scale to the plant scale; difficulty in correlating
geometallurgical variables with the variable of primary inter-
*Corresponding author, email peter.dowd@adelaide.edu.au est (usually grade); and the compositional nature of many

© 2016 The Australasian Institute of Mining and Metallurgy

Published by Taylor and Francis on behalf of the Institute of Materials, Minerals and Mining and The AusIMM Mining Technology   2016  vol. 125  no. 1  
Received 03 March 2015; accepted 29 August 2015
22 DOI 10.1179/1743286315Y.0000000032
 Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

1 
a Simulated copper grades in bench: three horizontal sections; b simulated copper grades in same bench: four vertical
sections; c blast profile resulting from simulated blast applied to simulated grades; d predicted composition of blast profile
from simulated blast applied to in situ grades estimated from samples taken from blast-holes on 8 m spacing

variables. Many of the deep mining resources are low-grade well-defined, geometrical shape. In practice, it is impossible,
complex ores, which require the spatial modelling and inte- for example, to blast a precise geometrical volume of ore from
gration of complex geometallurgical relationships. a larger volume such as a bench; blasting will inevitably result in
Our focus in this paper is largely on challenges (1), (2) and (3). ore loss and ore dilution, both of which may significantly affect
the production output.
Integrating orebody models with down- Here, for the sake of example, we demonstrate realistic quan-
tification of ore selection processes during mining. This requires
stream processes modelling and estimation of spatial rock breakage characteristics
In our view, the largest gap in current approaches to quanti- and spatial ore location indicators. In addition to providing more
tative strategic mine planning, design and optimisation is the accurate estimates of recovery, integrating the orebody model
need for realistic models of downstream processes to be applied with the selection process provides a means of optimising recov-
to orebody models. In other application areas, these processes ery, optimising blasting and loading strategies and optimising
are often termed transfer functions to emphasise that there is fragmentation to reduce energy costs at subsequent stages of the
an input (e.g., orebody model) to the process (e.g., a blast) and mining process (e.g., haulage, mineral processing). Analogous
an output (e.g., blast profile of fragmented ore and waste). The quantitative approaches can be devised for other processes and
integration of the sequence of inputs to, and outputs from, staged incorporated into models for each stage of the mining process.
processes is critical for mine optimisation and for optimising
the processes themselves. Orebody modelling, estimation and The method (Dowd and Dare-Bryan 2004) comprises:
simulation are now well developed, tested and widely used. But • generation of an in situ model of the orebody compris-
their integration into mining processes is less developed. Where ing the grade, geology, geomechanical properties and
orebody models are integrated with mining processes it is done, grade control variables within sufficiently small volumes
more often than not, in an over-simplified manner. For example, determined by the smallest selectable volume within a
the usual approach to modelling and estimating recoverability blast profile;
at different scales of selection is to assume perfect selection • definition of a blast volume comprising a large num-
of one well-defined geometrical shape from a larger, equally ber of the in situ model volumes, and subjecting it to

 Mining Technology  2016  vol. 125  no. 1   23

Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

2 
a Simulated (‘actual’) blast profile; b predicted blast profile; c ore zones in simulated (‘actual’) profile; d ore zones in
predicted blast profile

a blast simulator, which effectively moves each of the An example is given in Figure 1. Figure 1a and b show
component model volumes to its final resting place in simulated copper grades (‘reality’) for a mining bench of
the blast muck pile; and dimensions 80 × 40 × 10 m. The simulated bench is then sam-
• application of selective loading processes to the simu- pled on a specified scale, in this case 8 × 6 m blast holes and
lated muck pile to determine the degree of selectivity the blast hole sample grades are used to estimate the grades in
that can be achieved by various sizes of loader and the bench. The simulated (‘reality’) and the estimated grade
types of loading and to quantify ore dilution and ore volumes are then subjected to a blast simulator that generates
loss.The in situ model, representing perfect knowledge the ‘actual’ blast profile in Figure 1c and the predicted blast
at all relevant scales, is obtained by geostatistical sim- profile in Figure 1d. Selection and loading would be informed
ulation. An in situ model that represents the reality of and guided by the profile in Figure 1d, which would deliver
knowing only the data and information that are avail- a significantly different recovery than applying selection and
able from specific grade control drilling and sampling loading to the profile in Figure 1c; ore in Figure 1c is much
grids can be obtained by sampling the geostatistically more dispersed through the profile than in Figure 1d. The
simulated model on a specified grid. The volumes difference between selecting and loading from Figure 1c and
comprising the in situ model are then populated by d quantifies the ore loss and ore dilution for the amount and
estimates based only on the data corresponding to the location of sample data.
specified grade control drilling and blast-hole sampling To illustrate further the consequences of using the
grids. Different drilling and sampling grids can be used over-simplified geometrical shape approach to predicting
to generate different models, each reflecting the levels recoverability, consider the two blast profiles (from a different
of data and information available. Selectivity can then bench in the same orebody) in Figure 2. Figure 2a is the actual
be assessed as a function of the drilling and sampling blast profile and Figure 2b is the predicted blast profile based
grids as well as the size and type of loader. Performance on grades estimated from samples from a 8 × 6 m blast-hole
is assessed against the ideal selectivity achieved on the pattern. Figure 2c and d are the result of applying a cutoff
perfect knowledge model, comprising simulated values grade to define contiguous parcels of ore.
of each component volume. Applying costs, prices and Once the selection and recovery processes are quantified
financial criteria enables optimal selection of the grade and integrated with the orebody model, they can be used to
control drilling grid, size of loader, type of loading and plan, design and optimise the processes. For example, to
blast design.

24 Mining Technology   2016  vol. 125  no. 1

 Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

determine a blast design that will optimise fragmentation and composition and processing characteristics. The approach is
recovery for a specified budget. to optimise the system as a whole, rather than independently
optimising components of the system such as, for example,
optimally scheduling a block-grade model.
Geometallurgy Coward and Dowd (2014) summarise the current general
Geometallurgy originally referred to the incorporation of approach to geometallurgical modelling as:
metallurgical variables into spatial geological models of ore- • identify the variables required to understand critical
bodies to provide an integrated predictive basis for mine and process responses;
mineral processing design and optimisation. The definition • find ways to sample and measure these variables; and
has evolved over time to recognise the uncertainty of vari- • develop techniques to estimate and simulate these char-
ables and to extend the concept beyond the strict definition acteristics spatially at the correct scale and incorporate
of metallurgy; for example, ‘the integration of geological, the values into block models.The missing components
mining, metallurgical, environmental and economic infor- are the equivalent of the integrated transfer functions
mation to maximize the Net Present Value of an orebody or processes illustrated in the previous section, namely
while minimizing technical and operational risk’ (SGS 2014). to integrate the spatial geometallurgical model into a
Geometallurgy may be viewed as a specific example of the complete mine systems model to quantify the impact
integration of down-stream processing as covered in the pre- of variable and uncertain rock properties on all stages
vious section. However, the fundamental difference is that of process performance, mine design and optimisation.
geometallurgical variables are spatial variables that must be This aspect, with the addition of mining, environmen-
integrated in the block model whereas the processes covered tal and economic variables and processes, provides the
in the previous section are physical processes applied to the basis for an integrated systems approach to complex
block model (including geometallurgical variables). mining systems problems.
Ignoring the effects of geometallurgical variables in mine As an example, we provide some results from a geomet-
planning (and operation) disregards a critical section of the allurgical study of a polymetallic sulphide mineralisation in
value chain and, in many cases, leads to sub-optimal mine which surface alteration reduces progressively with depth
plans and operations. Authors take a systems approach to (Coward and Dowd 2014, 2015). The minerals of economic
mine planning and operation by, for example, optimising pro- interest are silver, lead, zinc, gold and copper. The oxida-
cessing routes for blocks of ore based on their mineralogical tion state, determined from core samples, is classified in six

Route 1

NSR1
Ex pit Leach
Ag Dore
Plant

Route 2
Conc 1
Float Plant

NSR2
For each block
Leach
Plant Ag Dore

Route 3
Conc 1
Float Plant
NSR3

Float Plant 2 Conc 2

Leach Ag Dore
Plant

Route 4
NSR4

Waste

3  Processing routes for a polymetallic sulphide deposit (Coward and Dowd, 2015)

 Mining Technology  2016  vol. 125  no. 1   25

Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

levels from fresh to extremely weathered. It is planned to of process uncertainty, normal distributions of errors were
leach weathered material and float fresh material. The spa- assumed and 25 alternative recovery values were simulated
tial block model comprises the five grade variables and one for each process and applied to the kriged block model (i.e.
geometallurgy variable in the form of a weathering index. The certainty assumed for the block model and uncertainty for the
weathering index is used to determine the processing route process models). The resulting combined metal in concentrate
for each block: direct leach; flotation followed by leaching curves are also shown in Figure 4.
tails; waste. Figure 3 shows a process flow chart with the four From the plots in Figure 4, it can be seen that in early years
possible routes, two of which have two or more sub-routes. of operation, the process and resource uncertainties overlap
Experimental values were used to establish recovery curves and there is very little difference in the two sets of curves.
for each metal for each process. An estimated block model Beyond year 13, the two sets of curves begin to deviate sig-
was created by kriging the values of the variables for each nificantly. The resource uncertainty curves beyond year 25
block. Values of the variables were also simulated at the block indicate that higher grades will have been mined and pro-
scale 100 times to provide 100 realisations of the orebody. cessed in the earlier years and production outputs become
In addition, a second estimated block model was created by more variable.
averaging the 100 simulations for each block. The expected This brief summary of an extended case study (Coward
recovery from each process was calculated from the multi- and Dowd 2014, 2015; Coward, Dowd and Vann 2013) is pre-
variate grades and recovery curves for each metal. sented here as an example of the integrated systems approach
For each of the 102 models, the net smelter return (NSR) to complete geometallurgical modelling. In the full case study,
was calculated for each route for each block. Any combination three different economic scenarios (different prices and costs)
of block and route that returns a positive NSR deems the block were also included. A more realistic approach would be to
to be ore for that route; otherwise the block is deemed to be include any interactions among the various uncertainties: in
waste. Each block was then assigned its optimal route together situ variables, processes and economic variables (and also
with the ranked remaining positive routes. An optimal pit was any environmental variables). In the example summarised
generated for each of the 102 models and the blocks within in Figure 4, this would have entailed simulating the process
each of these pits were scheduled by year for the mine-life of recoveries for each of the simulated block models. In the
the orebody. To simulate the operation, each individual block general case, for in situ geological and geometallurgical var-
is processed in sequence and, depending on the status of each iables, process models, economic variables, environmental
processing route, the block may or may not be sent through variables and the number of combinations and outcomes rap-
its optimal route. In cases where the optimal route is una- idly increases. Further significant increases in outcomes are
vailable (either due to being over-fed or to full downstream incurred if other process models, such as blasting, stockpiling
product stockpiles), the block is either stockpiled for later and blending are also included. The associated computational
processing (incurring a re-handling cost) or sent through an issues are examples of the fifth challenge and are visited again
alternative route if its contribution is positive and less than the in the section on IT Infrastructure and platforms.
re-handling cost. The process model thus reflects the actual The remaining issues for geometallurgical modelling are
decision-making that would occur during mine operation. identification and collection of sufficient data for metallur-
Schedules were generated in MineSight using the expected gical variables that are currently either not spatially sampled
NSR value and the qisk model was used to explore how or are grossly under sampled; spatial modelling of these var-
dynamic constraints might result in blocks not being treated iables and of their spatial relationships with primary varia-
through their optimal processing route, at times that may dif- bles such as grade; and, where relevant, adequate accounting
fer from the mine plan. for non-additivity and compositional constraints. The most
Figure 4 shows the resulting optimal tonnes of combined commonly quoted impediment in collecting and analysing
metal in concentrate for each year of the mine life for each of geometallurgical variables is cost. However, applications such
the 102 models and assuming no uncertainty in the process as the case study quoted here clearly demonstrate the added
recoveries. The variability in the 102 sets of plotted values is value that an integrated geometallurgical approach brings to
a measure of resource uncertainty. To provide an indication

4  Tonnes of metal in concentrate by year for simulated and estimated block models and simulated process recoveries. Two
sets of curves: 102 for resource uncertainty and 25 for process uncertainty. Up to year 12, there is very little difference
between the two sets of curves; beyond year 13, they begin to diverge significantly and by year 25 they are completely
distinct as indicated

26 Mining Technology   2016  vol. 125  no. 1

 Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

mine design, mine planning, scheduling and to optimising spacing of 20–25 m between the clusters; the smaller range
total system measures (such as NSR). gold variability is associated with the small range quartz var-
iability. Similar models and interpretations (with different
ranges) were found in the other principal geological direc-
Other types of variables tions; the complete analysis and resulting variogram models
can be found in Dowd (1995).
Many variables are not uniquely geological, geomechanical
This study enabled optimal blasting and selection strate-
or metallurgical in nature. For example, in an orebody in
gies and maximised the use of quartz occurrence as a proxy
which gold occurrence is associated with small quartz veins
ore location indicator. A specifically geometallurgical exten-
and rock fractures, 90% of the ore in any mining block may
sion would be to link crushing and grinding characteristics of
be contained in < 25% of the rock mass, which may not be
major rock types to the quartz indicator variable.
a single contiguous volume and there may be insufficient
data to identify the locations of veins and fractures. This has
important consequences for grade prediction, blast design, New models for new mining methods
rock fragmentation, selection, loading, crushing, grinding and
processing – and each of these processes cannot be considered The process models (or transfer functions) for traditional
independently of the others. mining methods are well known and, in many cases, the
Figure 5 shows a schematic north-south cross-section modelling methods (e.g., blast simulation) are readily avail-
through a low-grade, east-west striking gold orebody in able. However, as the industry transitions to different forms
an open-pit operation (Dowd 1995). Due to the low grade of mining there is a critical need to develop new models that
of the orebody most samples are from percussion drilling adequately characterise the physical processes and their asso-
and record gold in grams per tonne and quartz in percent- ciated or inherent uncertainties. The most significant exam-
age; fracture and vein information are lost in the percussion ples of new (either because of the scale and depth or the
chips. A statistical analysis shows insignificant correlation basic technology) mining methods are deep mining, particu-
between gold and quartz assays; it is more likely that the larly large-scale block caving, and solution mining (or in situ
correlation is between gold occurrence and the numbers of recovery). Deeper mining and in situ recovery will increase
quartz–granodiorite contacts, i.e. the numbers of veins and/ as current mineral resources are depleted.
or fractures, which cannot be determined from the samples.
A geostatistical analysis yields a north–south horizontal (i.e.
perpendicular to strike and dip) gold variogram with a nugget
Block caving
variance and two ranges: 4 and 20 m; and a quartz variogram There is an accelerating transition from surface mining to
with a nugget variance and a single range of 3 m. There is underground mining. The transition is being driven by cost as
no difference in the spatial variability of gold from samples well as access to deeper resources. According to Moss (2012),
that contain gold and quartz and those that contain only gold. block caving operating costs range from $US5 to $US7 per
Standard geostatistical analyses do not reveal any discernible tonne versus large open-pit operating costs of $US1.00 to
spatial co-variation between gold and quartz even though the $US1.20 per tonne. However, for the former, each tonne is
latter is the best available ore location indicator. A simple a tonne of ore whereas for the latter each tonne is a tonne of
analysis of the samples in each drill-hole shows that 90% of material moved.
all significant gold assays are either directly associated with As an indication of the rate of transition from surface to
quartz or are within 4 m of a quartz occurrence in the same underground mining, Rio Tinto expects 43% of its copper
hole; a down-hole distance of 4 m corresponds to a horizontal production and 25% of the entire world's copper production
distance of approximately 3 m. The variogram models are to come from underground operations by 2021 (Moss 2012).
consistent with the schematic model indicated in Figure 5: There is a significant increase in risk with large-scale cav-
veins and fractures occur in clusters of 3–5 m with an average ing not least because around 70% of the capital cost is incurred

5 Schematic cross-section of gold orebody with concepts of fracture and vein controlled mineralisation and associated ore
zones

 Mining Technology  2016  vol. 125  no. 1   27

Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

before any revenue is generated (Moss 2012). McCarthy retrieval of solutions post-mining will almost certainly
(2002) identifies two additional increased areas of risk asso- become an environmental protection requirement. Again, ade-
ciated with deep mining. First, the increased geological risk quate fracture and flow models will be critical to the ability
owing to sparse data and second, the increased technical risk to comply with such requirements.
in areas such as ground control, materials handling and safety.
To these could be added the additional types of data required
to characterise the rock mass for planning, designing and Other
managing large caves and the difficulty in obtaining direct
An additional area of application is in the conventional min-
measurements of the relevant variables. Brown (2012) pro-
ing of deep deposits, in which ground water invasion into
vides a comprehensive overview of the geotechnical risks
the mine and the environmental impact of mining on local
associated with caving.
aquifers will become critical. This requires tools for assessing
The fundamental scientific and technical challenge is leaching fluid and ground water movement as well as chemi-
to quantify and manage the rock mass behaviour under in cal reactions within the leaching field.
situ, pre-conditioning (hydraulic fracturing or blasting) and
caving conditions. In each of these phases, there are many
uncertainties to be quantified and included in quantitative Stochastic fracture models and fracture
process models. There is also a need to quantify and manage propagation modelling
geomechanical risk – departure from optimum performance
of the cave as well as ground control risk. The characterisation of rock fracture networks is a very diffi-
The basis for a quantitative model is a realistic fracture cult problem not least because accurate field measurement of
network and a fracture propagation model. For orebodies for a single fracture is difficult and measurement of all fractures
which large block sizes are likely to result from caving it is impossible. Thus, in practice, the whole fracture system
may be necessary to precondition by hydraulic fracturing. is not observable on any meaningful scale and the only real-
In designing fracture stimulation programmes, it is essential istic approach is via a stochastic model informed by sparse
to be able to predict fracture propagation with some level of data and/or by analogues. In block caving and solution min-
confidence. Ideally, the fracture stimulation processes should ing applications, a realistic solution is even more difficult
be designed to optimise pre-conditioning outcomes; the more as the only reference data related to the fracture system are
difficult challenge is to design them to maximise the ability from limited drill core samples, geophysical borehole logs
to control the cave or at least to predict the cave with an or sparse seismic events detected during hydraulic fracture
acceptable level of reliability. stimulation. For these reasons, most of the standard models
Other significant aspects of large-scale block caving are of fracture systems for these applications are oversimplified
material flow, dilution and draw-point blockages. Modelling and representations of reality.
prediction of all of these aspects requires a realistic model and Recently, however, there have been some significant
characterisation of the fractured (and fragmented) rock mass. advances in the stochastic modelling of fracture networks
and fracture propagation for hot dry rock geothermal res-
ervoirs. These developments are transferable, with minimal
Solution mining adaptation, to block caving and solution mining.
The immediate problem in any application is the require-
Solution mining, or in situ leaching or in situ recovery, in ment for sufficient data. Fracture networks in caving and solu-
which leaching solution is injected into stimulated fracture tion mining systems are essentially the product of hydraulic
networks and the dissolved mineral content is pumped to the fracture stimulations that, together with ground conditions and
surface, will become a more widely used mining method as the local stress regime, determine how fractures are formed
existing deposits are depleted and deeper and/or lower grade and propagated. Thus the data requirements are measurements
deposits, for which conventional mining methods are neither of in situ stress and fracture propagation.
technically nor economically feasible, are exploited. In 2011, The most obvious ways of collecting sufficient amounts of
45% of world uranium production was from in situ leaching fracture data are from geophysical sensing, measuring frac-
operations (World Nuclear Foundation 2014), an increase ture proxies and/or fracture propagation: acoustic impedance
from 16% in 2000. surveys, and recording of micro-seismic events triggered by
Solution mining requires a low-permeability reservoir to fracture stimulation during pre-conditioning tests, full-scale
be stimulated to create a connected fracture network to ena- pre-conditioning and/or production monitoring.
ble fluid flow. The fundamental problem is to induce flow The recent advances in fracture network modelling and
through the stimulated fracture network to achieve effective, propagation, using micro-seismic events triggered by fracture
efficient recovery at minimum cost and acceptably low envi- stimulation, can be broadly classified as stochastic optimi-
ronmental risk. Current models of flow through stimulated sation – optimal fitting of fracture shapes to clouds of seis-
fracture-based reservoirs are inadequate. The critical require- mic points. The methods developed include Markov Chain
ment is to develop a modelling technique that can solve large- Monte Carlo optimisation (Mardia et al. 2007; Xu, Dowd
scale problems efficiently without compromising on sufficient and Wyborn 2013), spatial clustering (Seifollahi et al. 2013),
fracture detail and fracture network connectivity. There are the Random Sampling Consensus (RANSAC) algorithm
many complex unresolved issues impeding a solution to this (Fadakar, Dowd and Xu 2013) and the Point And Surface
problem and one of the most critical is the lack of a clear Association Consensus (PANSAC) algorithm (Xu and Dowd
understanding of system behaviour, output and performance. 2014). All of these methods are available in a public domain
This is the direct consequence of the inability of standard software package (Xu and Dowd 2010). The methods can be
models to describe, effectively, realistically and efficiently, further classified into those that consider only the locations of
the fluid flow in a fracture-based reservoir on multiple scales. the seismic points and those that consider location and time
In situ leaching has significant potential owing to its low sequence of the events. The latter group provides the basis of
cost, low surface footprint and ‘greener’ operation. However, fracture propagation models.

28 Mining Technology   2016  vol. 125  no. 1

 Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

By way of example, we present here a short overview of

the PANSAC model together with outcomes from applying
it to the fracture stimulation of the Geodynamics’ Habanero
geothermal reservoir in the Cooper Basin. Full details of the
modelling methods and the specific application can be found
in Xu and Dowd (2014).
The fundamental assumption of the algorithm is that at
least one fracture passes through the location of any seismic
event detected. This is a reasonable assumption as seismic
events occurring in the fracture stimulation process result
from slips of existing fractures, creation of new fractures or
propagation of existing fractures. Fitting surfaces through
to a point cloud set is mathematically and computationally
difficult (Bercovier, Luzon and Pavlov 2002), particularly
when the fracture surface profiles are unknown and the data
points are sparse. A further widely used simplification is to
assume that, as a first-order approximation, planar polygons
can be used to represent actual tortuous fracture surfaces.
The RANSAC approach fits random fracture planes through
the point cloud by minimising a cost function. The resulting
fracture model consists of a sequence of fracture planes with 6  Absolute hypocentre locations of the seismic events
a number of associated points arranged in descending order,
implying the increasing uncertainty of fractures fitted at later
stages. RANSAC, and the earlier MCMC approach, ignore
the time sequence of the seismic events. Fracture stimulation
begins near the stimulation source and moves further away
from the source as the stimulation intensifies. In other words,
different parts of the rock masses are stimulated at different
times. As fracture creation or propagation is highly depend-
ent on any existing fracture network (Kear et al. 2013), it is
important to consider the sequence of fractures simulated at
earlier stages when fitting fracture models to a set of seismic
points. Failure to do so could potentially produce a misleading
fracture network model (Xu and Dowd 2014). The PANSAC
approach addresses this issue by taking every individual
seismic event in turn and optimising the association between
the point and the previously fitted fracture surfaces, taking
into account the existing fracture network, possible routes
(connections) back to the stimulation source and preferential
fracture propagation paths. The parameters used in the fitting
process are optimised using the fundamental principle that the
most likely fracture propagation path is the one that requires 7  Final PANSAC fracture model for the Habanero reservoir
the least amount of energy. The final fracture network fitted
by PANSAC consists of a sequence of fractures following
Figure 9 shows a sequence from the fracture propagation
closely the fracture stimulation process from the stimulation
model at different times in the stimulation process. The time
source to the boundary of the stimulation.
t is in units of days after 12:51pm, 30 October 2003, Central
Figures 6–9 illustrate the application of PANSAC
Australian time. Note that t from 12 to 18 corresponds to the
to Geodynamics’ Habanero geothermal reservoir in the
first major fracture propagation stage and t from 32 to 40
Cooper Basin of South Australia. The targeted hot granite
corresponds to the second major fracture propagation stage.
is about 4 km below the surface with a temperature close
The fitted fracture propagation model correctly captures these
to 250°C. Figure 6 shows the locations of 23 232 seismic
fracture propagation stages.
events detected during the fracture stimulation in 2003 using
This example clearly shows the potential for the application
Habanero 1 (H1) as the stimulation well. Eight geophones
of these models to characterising fracture networks and model-
deployed in wells surrounding H1 at depths of 100–2300 m
ling fracture propagation for block caving and solution mining.
were used to record the seismic signals generated by the
With the help of seismic monitoring during the pre-condition-
stimulation process. Note that wells H2 and H3 shown in
ing for block caving or ground stimulation for solution mining,
Figure 6 were drilled later and did not contribute to the events
these models can be used to assess the effectiveness of the
shown in the figure. The final fracture model generated by
stimulation process, the degree and the adequacy of ground
PANSAC is shown in Figure 7.
fracturing for the intended operations and to help optimise the
Figure 8a is a view of the fractures in Figure 7 looking
stimulation process to minimise the associated costs.
north and Figure 8b is the propagation sequence of the seismic
points projected onto an east-west vertical cross-section. Both
figures clearly show that a significant number of fractures dip Flow modelling for solution mining
gently to the West, that is the propagation of fractures trends
upward as they propagate towards the Eastern part of the The additional component for solution mining is to model the
reservoir. The PANSAC model correctly captures this feature. flow through the reservoir. Flow characteristics of a stimulated

 Mining Technology  2016  vol. 125  no. 1   29

Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

8  Dipping feature of the fractures in the East-west direction. a three-dimensional view; b North-south projection

fracture network depend mainly on the connectivity of the 35 L s(1 extraction rate. The temperature drawdown and power
fractures in the network. This connectivity can be represented produced can also be calculated accordingly.
by channels (connection pathways) connecting the injection The technique described here is readily applicable to solu-
and recovery wells through the fracture network. However, tion mining where the mine scale is, in general, significant
finding these channels is by no means simple, especially for and both flow and heat will play important roles in the min-
large size fracture networks. In our work, we search for all erals impregnation and transport processes. The additional
potential pathways and select on the basis of the shortest con- component that has to be considered in this application is the
nection paths but even so there are still significant numbers chemical reaction, which is relatively easier to model once
of potential flow pathways even for a moderate size fracture the flow and temperature models are established.
network. Figure 10 shows an example of the identified flow
channels through the fracture network between the H1 and
H3 wells for the MCMC model described in Xu et al. (2015) Flexibility in planning and design
. There are 27 293 channels in this case and most of these The ability to adapt mining operations to increasingly vola-
connection paths pass through more than 100 fracture planes. tile external environments (e.g., prices, markets) and chang-
Clearly, Habanero 1 is well connected to the surrounding rock ing internal circumstances (e.g., geomechanical conditions,
through the fracture model, which is expected because it is grades) is difficult but, nevertheless, critical for many pro-
the source of the fracture stimulation. Habanero 3 has only a jects. It requires sufficient flexibility in models to enable plans
small number of connections to the reservoir, which is also to be adapted and to manage risk and minimise its impact.
expected, as Habanero 3 plays no part in the fracture stimu- Models that attribute value to system flexibility (at strategic,
lation process. mid-term and tactical levels) will lead to plans and designs
For geothermal applications, once the detailed structure that can be adapted to deal with emerging risks and minimise
of the reservoir is modelled, further analysis in terms of fluid their impact on the operation.
flow and heat extraction is needed. For a large-scale reservoir The integrated model approaches advocated here provide
such as Habanero (approximately 2.0 × 2.0 × 0.5 km3) and with a basis on which flexibility can be valued. The integrated
detailed reservoir structures on the scale of metres, it is not systems modelling of variables, processes and their associated
possible to use conventional numerical tools due to the com- uncertainties allows options to be assessed, the system to be
putation cost (Bodin et al. 2007). The authors have developed re-optimised on the basis of changed circumstances, and risk
a simplified flow and heat extraction analysis tool, based on to be quantified and managed. In particular, the output for the
an equivalent pipe network, which significantly improves polymetallic sulphide case study given earlier in this paper
the computational efficiency and makes the reservoir scale is not only a fully flexible plan in terms of optimal process-
analysis possible (Xu et al. 2015). Figure 11 shows the pre- ing routes but also a fully flexible operational schedule that
dicted temperature distribution of the Habanero reservoir allows operators to choose the best available route when the
after 20 years of heat extraction based on H1–H3 doublet and optimal route is not available, i.e. the plan includes all relevant

30 Mining Technology   2016  vol. 125  no. 1

 Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

9  Fracture propagation model generated by PANSAC

 Mining Technology  2016  vol. 125  no. 1   31

Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

processing decision-making that would occur during actual The case study presented in this paper together with the
operations. somewhat limited overview of other work in this field is
The inclusion of quantified uncertainty in the modelling of intended to give an indication of the approaches that have
variables and processes during planning and design provides been successfully applied to real mining operations. A ques-
a sound basis on which to identify, prioritise and manage risk tion that arises is the extent to which these approaches have
and minimise its impact. As acknowledged by other authors, been adopted and meaningfully used in the mining indus-
(see, for example, Dimitrakoploulos et al. 2002), the quan- try. The question is particularly pertinent to the current eco-
tification of uncertainty and risk provides greater flexibility nomic conditions for most mined commodities – a situation
in mine planning and design and can add significant value for which the incorporation of uncertainty and flexibility in
to projects. mine planning and design is specifically intended.
The real options framework has been widely used to
accommodate uncertainty and flexibility in optimising mine
planning see, for example, Dimitrakopoulos et al., (2002), IT infrastructure and platforms
Dimitrakopoulos and Abdel Sabour (2007), Abdel Sabour There is a growing need for mining industry IT infrastructure
et al. (2008), Botin et al. (2012) and, in a broader context, and platforms for rapid on-line data collection, storage, access
Armstrong et al. (2004). Other authors have used more tradi- and processing and to support the increasing amount of auto-
tional methods such as Monte Carlo simulation of scenarios mation in many stages of the mining process. The increasing
(prices, costs, recoveries, etc.) together with optimisation generation and use of large sets of remotely sensed data at
methods (see, for example, Groeneveld and Topal 2011). mine and exploration sites require new platforms and infra-
Armstrong et al. (2013) also used real options to reduce the structure to realise the potential of the data in short-term deci-
number scenarios in project valuation, which is a significant sion-making and longer term strategic planning. These data
issue when using large numbers of simulations. range from down-hole sensing of mineralisation indicators
Another area of flexibility is in the physical design of and rock properties in deep exploration drilling to online sens-
mining operations. This work is best exemplified for under- ing of blast-hole cuttings and blast profiles to online sensing
ground 11mine design by the output of Brazil et al. (2003, of truck performance. The effective use of these data requires
2004, 2009). immediate, and often remote, access to very large data sets
and rapid processing of data to optimise processes such as
drilling directions, blast design, selective loading and choice
of mineral processing routes.
While the largest companies have established their own
facilities, it makes sense for others to come to collaborative
arrangements for joint provision with appropriate protocols
to ensure security of commercial data and information. Given
the key economic role that mining plays in many economies,
governments may take the view that this is essential infra-
structure and provide support in establishing it.
The Collaborative Remote Operations Centre project
funded by the South Australian State Government's Mining
and Petroleum Services Centre of Excellence (DMITRE
2014) is a good example of an industry/government/uni-
versity collaboration to provide an IT platform to support
automation, remote analysis, data storage and on-line access,
and the testing of software and hardware for remote mining
operations. The collaborative qualifier refers to the shared
nature of the Centre.
10  Connection channels between Habanero 1 and 3

11  Reservoir temperature distribution after 20 years production

32 Mining Technology   2016  vol. 125  no. 1

 Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

Collaborative centres such as this require ICT architectures Armstrong, M., Ndiaye, A., Razanatsimba, R. and Galli, A. 2013 Scenario
that provide rigorously secure access to the IT systems of reduction applied to geostatistical simulations. Mathematical Geosciences,
45, (2), 165–182.
members so as to allow data integration and analysis on the Armstrong, A., Galli, A., and Couët, B. 2004. Incorporating technical
platform without compromising the security of a member's uncertainty in real option valuation of oil projects, Journal of Petroleum
data. The University of South Australia, in collaboration with Science and Engineering, 44, (1-2), 67–82..
mining company OZ Minerals Ltd, has led the development Bercovier, M., Luzon, M. and Pavlov, E. 2002. Detecting planar patches
in an unorganized set of points in space, Advances in Computational
of a ‘test and trial’ Collaborative Remote Operations Centre, Mathematics, 17, 153–166.
that is intended to provide a future innovation hub for the min- Bodin, J., Porel, G., Delay, F., Ubertosi, F., Bernard, S. and de Dreuzy, J.-
ing industry with longer term access for students, researchers R. 2007. Simulation and analysis of solute transport in 2D fracture/pipe
and ICT companies for research, teaching and testing of new networks: The SOLFRAC program, Journal of Contaminant Hydrology,
products and services (Sriram et al. 2014). 89, 1–28.
Botin, J. A., Dell Castillo, M. F., Guzmán, R. R. and Smith, L. 2012. Real
New real-time logging while drilling (LWD) technologies options: a tool for managing technical risk in a mine plan. In: SME Annual
are being developed in the Deep Exploration Technologies Meeting. Seattle, USA, SME, Pre-print 12–121, 7.
Co-operative Research Centre (DETCRC 2015). Almost all mine Brazil, M., Grossman, P. A., Lee, D., Rubinstein, J. H., Thomas, D. A. and
design and planning are based on analyses of core samples and Wormald, N. C. 2009. Access Optimisation tools in underground mine
design. International Symposium on Orebody Modelling and Strategic
hence the modelling and optimisation methods used to deliver Mine Planning: old and new dimensions in a changing world, 237–241,
designs, plans and schedules are, of necessity, largely limited to Melbourne, Australasian Institute of Mining and Metallurgy.
grades and tonnages. However, cores limit the types of data that Brazil, M., Lee, D. H., Van Leuven, M., Rubinstein, J. H., Thomas, D. A.
can be collected and they reduce productivity because of time and Wormald, N. C. 2003. Optimising declines in underground mines,
lost in transporting them to laboratories and analysing them. In Transactions of the Institution of Mining and Metallurgy, Section A: Mining
Technology, 112, 164–170.
addition, cores limit the type of drilling technologies, which is Brazil, M., Lee, D., Rubinstein, J. H., Thomas, D. A., Weng, J. F. and Wormald,
one of the focus areas of DETCRC. The potential of LWD for N. C. 2004. Optimisation in the design of underground mine access. In:
mine planning and design is real-time information on geological Orebody Modelling and Strategic Mine Planning, Spectrum Series Volume
controls on mineralisation, delineation of ore zones, geotechni- 14. Australasian Institute of Mining and Metallurgy, Melbourne, Australia,
121–124.
cal characteristics and, possibly, geometallurgical characteristics Brown, E. T. 2012. Progress and challenges in some areas of deep mining,
particularly those related to rock breakage. One of the key objec- Transactions of the Institution of Mining and Metallurgy, Section A: Mining
tives of programme 2 of DETCRC is to develop technologies Technology, 121, 177–191.
to invert LWD-sensed petrophysical data for rapid updating of Coward, S. and Dowd, P. A. 2014. Geometallurgical models and the
three-dimensional geological models. In the deep exploration quantification of uncertainty in mining project value chains. In:
Presentation at Geometallurgy 2014, London, Institute of Materials,
context, these very large LWD data sets are already available and Minerals and Mining.
similar types of data are being generated in operating mines. In Coward, S. and Dowd, P. A. 2015. Geometallurgical models for the
the context of strategic mine planning, the challenge is to adapt quantification of uncertainty in mining project value chains. In:
modelling, design, estimation, simulation and optimisation algo- Proceedings of the 37th APCOM Conference. publication of Society for
Mining, Metallurgy and Exploration (SME), 360–369.
rithms to real-time processing of these types of rapidly acquired, Coward, S., Dowd, P. A. and Vann, J. 2013. Value chain modelling to evaluate
and very large, data sets so as to improve the speed and quality geometallurgical recovery factors. In: Proceedings of the 36th APCOM
of decision-making. Conference, 288–289, Brazil, pub. Fundação Luiz Englert.
DETCRC. 2015. Available at: < http://detcrc.com.au>.
Dimitrakopoulos, R. G. and Abdel Sabour, S. A. 2007. Evaluating mine plans
Conclusions under uncertainty: can real options make a difference?, Resources Policy,
32, 116–125.
Most of the challenges outlined here do not require new models Dimitrakopoulos, R., Farrelly, C. T. and Godoy, M. 2002. Moving forward
or new approaches, they require adapting existing techniques to from traditional optimization: grade uncertainty and risk effects in open-pit
new applications (as in fracture modelling for caving and frac- design, Transactions of the Institution of Mining and Metallurgy, Section
A: Mining Technology, 111, (1), 82–88.
ture network and fluid flow modelling for in situ leaching) and DMITRE (Department for Manufacturing, Innovation, Trade, Resources and
a systems approach to integrating resource/reserve models with Energy). 2014. Available at: < http://www.dmitre.sa.gov.au/manufacturing_
mine processes (as in geometallurgy and the more general mining works/programs_and_initiatives/mining_industry_participation_office_
response or transfer functions). There is a need for more work mipo/mining_and_petroleum_services_centre_of_excellence>.
Dowd, P. A. 1995. Björkdal gold-mining project, northern Sweden,
on adequate spatial modelling of geometallurgical variables, par- Transactions - Institution of Mining and Metallurgy, Section A: Mining
ticularly in dealing with non-additive and compositional varia- Industry, 104, A149–A163.
bles, and in providing tested and validated response functions. Dowd, P. A. and Dare-Bryan, P. C. 2004. Planning, designing and optimising
There is also a need for the computational speed of modelling, using geostatistical simulation, In: Proceedings of the International
simulation and estimation algorithms to be commensurate with Symposium on Orebody Modelling and Strategic Mine Planning. Pub
AusIMM (Melbourne), Vol. 22, 9, 321–338.
the speed with which sensed data are collected and decisions on Fadakar, A. Y., Dowd, P. A. and Xu, C. 2013. The RANSAC method for
them need to be made. generating fracture networks from micro-seismic event data, Mathematical
In the context of the integrated systems approach and the Geosciences, 45, 207–224.
need for flexibility in models and outcomes proposed here it Groeneveld, B. and Topal, E. 2011. Flexible open-pit mine design under
uncertainty, Journal of Mining Science, 47, (2), 212–226.
may be more important to ensure that outcomes are robust and Kear, J., White, J., Bunger, A. P., Jeffrey, R. and Hessami, M. A. 2013.
resilient rather than strictly ‘optimal’ for a rigidly specified Proceedings of International Conference for Effective and Sustainable
set of geological, geometallurgical, operational and economic Hydraulic Fracturing 2013, Brisbane.
criteria. Mardia, K. V., Nyirongo, V. B., Walder, A. N., Xu, C., Dowd, P. A., Fowell,
R. J. and Kent, J. T. 2007. Markov Chain Monte Carlo implementation of
rock fracture modelling, Mathematical Geology, 39, 355–381.
McCarthy, P. 2002. Flexible studies and economic models for deep mines: In
References Proceedings of the First International Seminar on Deep and High Stress
Mining, Perth, Australian Centre for Geomechanics.
Abdel Sabour, S. A., Dimitrakopoulos, R. and Kumral, M. 2008. Mine design Moss, A. 2012. Keynote address at MassMin 2012, Sudbury, Canada.
selection under uncertainty, Transactions of the Institution of Mining and [unpublished].
Metallurgy, Section A: Mining Technology, 117, (2), 53–64.

 Mining Technology  2016  vol. 125  no. 1   33

Dowd et al.  Strategic mine planning and design: some challenges and strategies for addressing them

Seifollahi, S., Dowd, P. A., Xu, C. and Fadakar-Alghalandis, Y. 2013. A Xu, C. and Dowd, P. A. 2014. Stochastic fracture propagation modelling
spatial clustering approach for stochastic fracture network modelling, for enhanced geothermal systems, Mathematical Geosciences, 46, (6),
Rock Mechanics and Rock Engineering, 47, (4), 1225–1235. 665–690.
SGS. 2014. Available at: < http://www.sgs.com> [accessed 12 June, 2014]. Xu, C. and Dowd, P. A. 2015. A simplified coupled hydrothermal model for
Sriram, V., Kearney, D. and Andrews, S. 2014. Collaborative Remote enhanced geothermal systems, Applied Energy, 140, 135–145.
Operations Centre Report, Department of State Development, Government Xu, C.Dowd, P. A. and Fidelibus, C. 2014. Realistic pipe models for flow
of South Australia, Available at: < http://www.statedevelopment.sa.gov. modelling in Discrete Fracture Networks. In: Proceedings of the First
au/upload/manufacturing/1446_collaborative_remote_centre_operations_ International Discrete Fracture Network Engineering Conference.
report.pdf>. Vancouver, Canada.
World Nuclear Foundation. 2014. Available at: < http://www.world-nuclear. Xu, C., Dowd, P. A. and Wyborn, D. 2013. Optimisation of a stochastic rock
org/info/Nuclear-Fuel-Cycle/Mining-of-Uranium/In-Situ-Leach-Mining- fracture model using Markov Chain Monte Carlo simulation, Mining
of-Uranium> [Accessed 28 May, 2014]. Technology, 122, (3), 153–158.
Xu, C. and Dowd, P. A. 2010. A new computer code for discrete fracture
network modelling, Computers and Geosciences, 36, 292–301.

34 Mining Technology   2016  vol. 125  no. 1

Copyright of Mining Technology is the property of Taylor & Francis Ltd and its content may
not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for
individual use.