International Journal of Rock Mechanics & Mining Sciences 178 (2024) 105778

Contents lists available at ScienceDirect

International Journal of Rock Mechanics and Mining Sciences

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

Optimising stope design through economic and geotechnic assessments of

predictions made at a meter scale resolution using the sites’ reconciled data
Benoît McFadyen a, Martin Grenon a, *, Kyle Woodward b, Yves Potvin b
a
Université Laval, Quebec City, Quebec, G1V 0A6, Canada
b
Australian Centre for Geomechanics, The University of Western Australia, Perth, WA, 6009, Australia

A R T I C L E I N F O A B S T R A C T

Keywords: The final geometry of open stopes often diverges from the design, creating unintentional volumes of overbreak
Stope design and underbreak. Tools, such as the Stability Chart, are used for predicting overbreak, but its limitations regarding
Stope reconciliation the precision of the predictions and the exclusion of underbreak have constrained the benefits of a predictive tool
Overbreak
at the design stage. Furthermore, the economic aspects are not integrated when addressing geomechanical stope
Underbreak
Prediction
design. This paper presents how understanding and predicting stope performance at a meter-scale resolution
Economic through a multivariate model can be used for assessing the geotechnical (overbreak and underbreak) and eco­
nomic stope performance. These predictions estimate the magnitude and location of overbreak and underbreak
across the design surface, and are used to resolve the expected geometry of the mined void. The incorporation of
a predicted mined void with grade values from a block model allows for a comprehensive economic stope
performance estimate. This work is a significant step towards a stope design and planning process that minimizes
the stability issues, dilution and loss of ore across the stope. As designing for higher ore recovery increases the
potential for dilution and vice versa, this approach allows for the value realised from mining a stope to be
maximised by quantifying this inherent economic trade-off. The enhanced resolution of both data and predictions
has allowed for a more comprehensive evaluation of the predicted stope performance and economic outcomes
during the design phase. This marks a notable advancement in the ability to design and plan the optimal stope.

1. Introduction A popular predictive tool used in stope design is the Stability Chart2
or one of its subsequent versions. This empirical tool looks at OB,
Open stoping is a popular mining method due to the high production enabling to qualitatively predict whether a stope surface will be stable,
levels that can be achieved.1 To maximize the profitability of the oper­ unstable or caving. The stability is assessed by calculating a stability
ation, one should maximize the volume of ore per stope while respecting number using the geomechanical properties of the rock mass and
the designed geometry when mining it. However, what often occurs calculating the hydraulic radius using the planned stope geometry. The
when mining is the generation of volumes of overbreak (OB; rock or tool is used to make decisions on the planned geometry or the need for
backfill material outside of the design boundaries) or underbreak (UB; ground support. In the absence of stope performance and operational
rock inside the designed boundaries left unrecovered). These unin­ data, the Stability Chart is a valuable tool at the feasibility study stage.
tended volumes can have a critical impact on the operation and the Once mining has started and extraction progresses, operational and
profitability of the mine. OB can cause stability issues that affect adja­ stope performance data become available. However, there is no efficient
cent stopes and mining infrastructure, and delay the mucking process. It process to provide on operational or quantitative stability performance
can also dilute the ore with waste material that can cause non-optimal data to the Stability Chart. This severely limits the optimisation capa­
recovery at the mill. UB, on the other hand, can result in a loss of ore bility of the method. Furthermore, the Stability Chart does not consider
as it is left behind and can’t be recovered. In order to minimise these OB UB which can have a major impact on the revenue realised from mining
and UB volumes, engineers will look to understand and predict these OB a stope or the profitability of a stope. These limitations inherent to the
and UB volumes. Stability Chart decrease its utility when aiming to predict stope

* Corresponding author.
E-mail addresses: [email protected] (B. McFadyen), [email protected] (M. Grenon), [email protected] (K. Woodward), yves.
[email protected] (Y. Potvin).

https://doi.org/10.1016/j.ijrmms.2024.105778
Received 26 October 2023; Received in revised form 16 May 2024; Accepted 18 May 2024
Available online 27 May 2024
1365-1609/© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

stability and economics in silos prevents a more-efficient stope design

optimisation.
The limitations to predicting stope performance and the exclusion of
economic data highlight the need in the mining industry for developing
a new empirical approach to stope design. An approach that can
incorporate the different parameters critical to stope performance may
enable mines to understand and quantitatively predict stope perfor­
mance (OB, UB and grade) at the design stage to assess and optimise
stope stability and profitability. This novel approach can be achieved
through statistical multivariate analysis4,5 and stope reconciliation at a
significantly finer resolution than on a per face basis.6
This paper presents how to conduct an integrated economic and root
cause assessment of the predicted OB and UB and their associated
probabilities, obtained using multivariate statistics and data measured
at an approximately cubic meter resolution, in order to work towards a
more-optimal stope design. This new methodology presents many nov­
elties facilitated by the increased resolution of the data and the inclusion
of a wide range of parameters impacting stope performance. The data
structure follows an octree representation and provides georeferenced
points along the stope design surface. This structure provides a fine
Fig. 1. Stope overbreak and underbreak with respect to a stoping min­ spatial reference stope OB and UB compared to a per face basis. A
ing method14. detailed economic reconciliation can be achieved by spatially mapping a
grade block model to the octree data structure built from predicted ge­
ometry, which integrates the financial aspect of mining into the
Table 1
Parameters quantified on a per face basis.
decision-making process of stope design. This research represents a
significant leap in the capability of mine sites to predict stope perfor­
Parameter Description Reference and
mance (OB, UB and profit) at the design stage. Economic-based opti­
example of
application misation represents the final step toward a new stope design
methodology which can be achieved if the predictive model considers
Geometrical:
Hydraulic radius Ratio between the area of a face 2
parameters that can be adjusted during stope design and if economic
and its perimeter. data is quantified. The following work details the approach for assessing
Dip Average angle with respect to the 2 and using the predictions in the stope design step and presents the re­
vertical axis for a stope face (◦ ). 5 sults of a case study where the novel approach was tested.
Operational:
Blasting standoff Average Distance of the closest 9
drill ring to the face (m). 34 2. Background on stope optimisation
5
Blasting orientation Average angle between the 9 Optimal stope design can be achieved by building an understanding
boreholes and the face for the 34 on the root causes of OB and UB, and integrating this knowledge with
closest ring (◦ ). 5
Undercut Width of the drive exceeding the 9
economic data at the design stage so that engineers can utilise this key
stope boundary (m). 10 insight to optimise or plan stope extraction and meet the geotechnical
Ground support Presence of cables on the stope 35 and economic goals set by mine management. This knowledge is gained
face, quantified as bolt density, 36 by analysing and understanding the mine site’s historic stope perfor­
cable length and unsupported
mance. Some stope performance issues can be managed with changes
span.
Stand-up time Number of days between the final 10 that are site wide or per zone, such as changing the mining sequence for
blast and the final CMS. 37, managing the induced stress. However, some stope performance issues
5 are localised and will vary from stope to stope or within the stope. In
Geomechanical: order to assess the expected performance of a designed stope, sites turn
Classification of the Rock mass quality according to 2,
rock mass (Q′, RQD, one of the standard 5
towards predictive tools that will quantify the stope OB and UB. The
RMR, etc.) classifications. predictions can then be assessed to determine the paths for planning or
Stress regime (σ1, σ2, σ3 Characterisation of the induced 2 optimising the stope. The key steps for achieving an optimal design are
etc.) stress and deformation around 38 stope reconciliation, the prediction of stope performance and the eco­
the stope. 39
nomic analysis.
Geological:
Presence of faults Position, angle and Distance of a 40
fault to a face. 41 2.1. Stope reconciliation

Stope OB and UB are quantified by reconciling the design geometry

performance and optimise stope design. and the mined geometry (Fig. 1) obtained with a cavity monitoring
Site-specific tools can be developed using the previously mined stope
system (CMS) scan7 or, as of late, a LiDAR mounted on a drone.8 The
data identifying parameters that are the root causes of OB and UB. This most common levels of resolution are on a per stope and per face basis.3
knowledge can then be used at the design stage. However, workshops
Stope performance is quantified as volumes or percentage, and, for faces
conducted in 2019 with mine sites in Canada and Australia have shown only, as equivalent linear overbreak slough (ELOS,9) and equivalent
that mine sites produce limited information from their reconciliation
linear lost ore (ELLO).
even though a large quantity of data is being collected.3 This results in During the reconciliation process, parameters are quantified for the
limited information and knowledge being fed back into the design of
root cause assessment and reflect the general understanding of rock
future stopes. Furthermore, stope economics were not considered when mechanics that control OB and UB. The parameters will vary according
addressing OB and UB from a stability perspective. Assessing stope
to the resolution and the available mine site data. A limited number of

2
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Fig. 2. Illustration of the recursive process of octrees as well as the octree stope performance quantified by calculating the Distance in a direction normal to the
design surface between the design surface and the CMS for each octree (modified from33).

parameters is quantified on a per stope basis (e.g. design volume and outputs a qualitative or quantitative measure of OB or UB. These tools
mining method). More parameters can be quantified per face (Table 1). are empirical approaches and rely on the mining experience as well as
However, these parameters usually represent an average value for the general concepts of rock mechanics for predicting stope performance.
face. The predictive tools can be used at different stages of the design to
There is also a finer resolution that represent a new advancement in determine the expected stability of the stopes and plan or optimise. The
stope reconciliation where the performance is assessed on a per octree first tool that was developed is the Stability Chart by Matthews et al., in
basis.6 The three-dimensional space is recursively subdivided into blocks 1981. It allows, at the feasibility stage to determine the general OB and
(octrees) to a given size along the design surface that is usually less than select stope dimensions to come up with an initial optimal geometry.
a meter (Fig. 2). At this resolution, thousands of data points are gener­ The chart uses fixed parameters (stability number and hydraulic radius)
ated along the design surface, exponentially increasing the amount of to determine if the face OB will be stable, unstable or caving. As
data points compared to a per face basis. Stope performance at an octree mentioned in the introduction, the chart presents several limitations
resolution is quantified for each octree by calculating the Distance from regarding the predictions, namely, the precision of predicted OB and
the design surface to the CMS surface in the direction normal to the exclusion of UB, and the difficulty to include new factors such as drilling
design surface (Fig. 2). This Distance is referred to as the projected and blasting in the chart’s structure. Furthermore, a bivariate approach
Distance. If this measure is positive or negative, projected distances are is limited as parameters can be cross correlated and interdependent.
classified as OB and UB. The shortcomings of the Stability Chart have led to the exploration of
Given the finer resolution of the octree data structure, geo­ statistical models for predicting OB or UB that offer a flexibility for
mechanical, geometrical, geological and operational parameters can be integrating other parameters. These models can be used once mining has
refined.6 The resolution enables the spatial variation of values to be started as they rely on the site’s previously mined stopes. Multiple
captured along the design surface, addressing the loss of information multivariate statistical models have been tested and used for classifying
caused by averaging values over a stope face and increasing the resolute or quantifying OB on a per face basis. These models are listed below:
of analysis that can be achieved.
• Multiple linear regression.4,10,11
2.2. Predicting stope performance • Logistic regression and principal component logistic regression.4
• Partial least squares.5
Predictive tools are built on the understanding of the root causes of • Random forest.12
OB and UB which is done through analysis of the quantified parameters. • Artificial neural network.13
The identified critical parameters feed into the charts or models which
These models showed accurate predictions or classifications of OB on
a per face basis and use additional critical parameters making them
more flexible and site-specific approaches. They can also be more pre­
cise than most of the stability charts by predicting ELOS or ELLO in
comparison to general stability state (stable, unstable, caving). The
statistical models quantify directly (correlation coefficients) or indi­
rectly (models’ variable importance) the impact of each parameter on
the predictions and can be used to assess the root causes of the predicted
stope performance. The main limitation with these models is that
separate models are needed for OB and UB, and their implementation
are on a per face basis, thus not enabling the assessment of stope per­
formance at a finer resolution. Multivariate statistic models (partial least
squares, linear discriminant analysis and random forest) at an octree
resolution can be applied to address these limitations.14 The resolution
of the data enables predicting the magnitude and location of OB and UB
in one model which allows a predicted mined geometry to be con­
Fig. 3. The proposed four steps for assessing the predicted stope performance structed from these results.
as part of the stope design methodology. While these models and charts will provide a prediction for a set of

3
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Fig. 4. An example of how stope performance can spatially vary along a stope surface33.

Fig. 5. Example of a) different prediction geometries for a stope design shape (black) according to probabilities. The blue, green and red surfaces are the 20th, 50th
and 80th percentile cases of the prediction interval, respectively; b) of the predicted Distance for an octree based on the CDF and percentile considered. (For
interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

input values, there can be uncertainties and variability in the parameters risk analysis. It is also used by17 to consider the effect of the variability in
that go into these tools due to the rock mass sampling, the rock mass the geomechanical parameters for a numerical model in order to
heterogeneity or the data quality. This implies possible prediction er­ determine the optimal stope dimensions. However, this probabilistic
rors. Suorineni et al.15 showed using likelihood statistics the probability approach relies on the distribution of the parameters being known. It is
for a prediction using the stability chart to be wrong. To encompass also only applied on the predictions and does the not consider the
these factors, a probabilistic approach can be applied to provide a pre­ models or the chart variance.
diction interval in order to evaluate the probability of large OB or UB.
This approach relies on quantifying the distribution of each input 2.3. Economic analysis
parameter to consider the variability. This approach is used by16 with
the Stability Chart to quantify the stability interval in order to perform a Stope profitability can be complex to assess as it will depend on

4
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Fig. 6. Example of the calculation of the NSR value per octree (left) and an example of the result per octree block of the NSR value adjusted with backfill infor­
mation (right).

Fig. 7. Example of a) the cumulative distribution of the NSR value for each tonne based on the different categories of volume. Total tonnes per volume are indicated
as a footnote and the cut-off value with the black line; b) 3D view of the NSR value for the OB octrees. Points with an NSR value below the selected cut-off grade are
semitransparent while points above the cut-off grade are opaque.

different mining costs, rate of mining, amount of metal in the stope, sale and treatment costs at the mill. The NSR considers all the metals being
price, optimal smelter feedstock, ore recovery, fragmentation, dilution, mined and is calculated on a point basis along with the grades of all the
lost of ore and more. To simplify the stope design and calculate its metals in that location. The grade and NSR values can be presented as a
profitability, mines will simplify part of the economic analysis by block model and used with the cut-off grade to determine the limits for
determining the cut-off grade, and, when possible, the net smelting re­ the design of the stopes.19 Since OB and UB inevitability occur when
turn (NSR). mining a stope, the effect of dilution and loss of ore must be considered
The cut-off grade is the minimum grade of ore that is economically to assess the profitability of the stope. The loss of ore represents the
viable to mine and is calculated, depending on the metal, as a percentage amount of ore left in the stope. The dilution represents the amount of
or grams per tonne and can be expressed as dollars per tonne if multiple waste that is mixed with the ore (Equation (1),20). Dilution can be
metals are mined. The cut-off grade represents the limit between ore and planned (within the design) or unplanned (from OB) and will incur
waste. The calculation of the cut-off grade will depend on different direct (blasting, mucking, crushing etc) and indirect (metal recoveries)
factors, such as recovery, sale price, mining costs, processing cost, ca­ costs.21 The level of tolerance to dilution and loss of ore will depend on
pacity of the mill and refinery, discount rate and so forth.18 Some of the grade and quantity of ore in the stope, OB and UB, and will vary from
these factors will vary spatially and through time, thus the cut-off grade mine to mine. For example, dilution for a mine with a disseminated
will also vary. orebody will still contain metals and will not influence the overall grade
The NSR represents the estimated value of the ore, calculated in to the same degree as a mine with a distinct separation between ore and
$/tonne, and is based on the sale price of the metals minus the transport host rock.

5
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Table 2 prediction assessment process is described in four steps (Fig. 3).

Different dilution and loss of ore parameters.
Parameters Formula 3.1. Octree data
20
Predicted dilution Waste in mined
Ore in mined The data used in this new stope design approach is quantified using
Planned dilution42 Waste in design an octree data structure at a meter-scale resolution. The two main rea­
Ore in design
Unplanned dilution42 Waste in OB
sons are the increased spatial information and parametric possibilities
Ore in design enabled by this resolution. Since information is quantified on a point
Loss of ore21 Ore in UB basis along the design surface and at a meter-scale resolution, the stope
Ore in design performance spatial variation across the stope design is characterised.
Recovered ore Ore in mined
An example is given in Fig. 4 where the variation from OB in the left to
Ore in design
the UB on the right of the hanging wall is characterised through thou­
sands of georeferenced octrees.
Waste tonnes mined Parameters are quantified for root cause analysis for each octree
Dilution =
Ore tonnes mined location. Given the resolution, geometrical, operational, geomechanical
Or (1) and geological parameters can be quantified, enabling to understand
Waste tonnes mined why there is significant OB on the left of the hanging wall and UB on the
Dilution = right in Fig. 4. The increased resolution also enables to quantify pa­
Ore tonnes mined + waste tonnes mined
rameters that couldn’t be quantified before at a per face resolution;
Dilution is managed through strategic (mining method chosen, stope therefore, offering a wider range of parameters as well as the possibility
design) and tactical (cable bolt, blasting, backfill, draw control) con­ to develop a predictive model for optimisation. The list of parameters
trols.22 Urli and Esmaieli23 showed the impact of using an ore-skin considered in the new stope design approach have been presented in.24
design on the hangingwall dilution and its cost benefit. If dilution and The parameters are quantified using the scripts developed in the
ore loss can be predicted, it is possible to be proactive with these man­ mXrap25 stope reconciliation and analysis app.26
agement methods. Predicting the grade value at the design stage using
current design tools is limited as an average value (ELOS) or a general
3.2. Predictive model
stability level (stable, unstable or caving) for a stope surface is predicted
while UB is excluded. Dilution and ore loss can be estimated if the ELOS
A multivariate statistical model is used to predict stope performance.
or ELLO is predicted, but only if there is a clear boundary between ore
Given the use of octree data, the magnitude of OB or UB (projected
and waste, with the design surface falling on this boundary. Therefore,
distance) is predicted on a point basis enabling to predict the expected
the average prediction per face inhibits the accurate quantification of
mined geometry. The statistical model used for predicting the stope
the grade value of the OB and UB as well as operational optimisation.
performance is random forest.27 The statistical approach is selected
Predictions at a meter-scale resolution would, however, enable the
based on the work done by,14 showing strong prediction performance at
quantification of the grade value. The next section will detail the
a meter-scale resolution. Random forest is a tree-based non-linear
methodology for predicting and assessing stope performance at a
approach that can consider a complex dataset and predict with high
meter-scale resolution.
accuracy. Random forest enables to build an empirical predictive model
using the quantified critical parameters as well as being adaptable to
3. Methodology
each mine site. The predictive models are built using the Ranger script28
in the R software.29 The predicted geometry is built using the Poisson
In this article, a new stope design approach is proposed that uses
surface reconstruction process30 in CloudCompare.31
reconciliation data and prediction at a fine resolution to predict and
assess stope performance (OB, UB and grade). This approach enables one
3.2.1. Probabilities
to work towards a method to optimise stope design. This article will
The use of a random forest predictive model enables to building a
demonstrate this new stope design methodology using a case study. The
probabilistic design approach. A prediction obtained using random forest

Fig. 8. 3D section view of the stope database.

6
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Table 3
Critical parameters established through root cause analysis with a description (modified from6,24).
Parameter Concept Interpretation Example

Blasthole Distance (coloured by meter) from an octree block to the Blast-induced damage.
standoff nearest designed blasthole. Characterises how OB and UB are
distance. affected by the standoff distance to
the blastholes.

Distribution of the Distance for each octree along the design

surface (dots in the image) to their nearest blasthole. Cold
colours mean the octrees are further away from the blastholes
and generally, UB is expected.
Blasthole Shortest angle (coloured by degree) between an octree’s Blast-induced damage.
orientation. design normal vector and the blasthole. Convention: 90◦ Characterises how OB and UB is
= Toeing, 0◦ = Parallel. affected by the angle of the
blastholes to the design surface.

Distribution of the angle for each octree along the design

surface (dots in the image) to their nearest blasthole. Warm
colours (90◦ ) means the blastholes are toeing into the design
surface.
Proxy for blast ∑n L Blast-induced damage.
Blast energy ≈ 0 where:
energy. max (0.5, D)2 Characterises how OB and UB are
L = Length of hole. affected by the density of blastholes
D = Distance to hole. and the drill pattern.
Max. distance of 0.5 m to account for stemming near
collars (coloured by energy index).

Distribution of the energy proxy for each octree along the

design surface (dots in the image). The energy calculation
considers the density of blastholes around the octree (inverse
distance relationship) and the length of the blastholes to
determine the spatial distribution of the blasting energy
according to the blasthole design. Warm colours mean more
energy is expected in that area due to the drill design.
Dip of the The dip calculated from the normal vector associated Influence of gravity.
design with the octree block (coloured by degree). Characterises how OB and UB are
surface. affected by the effect of gravity.

Distribution of the dip for each octree along the design surface
(dots in the image). Warm colours (90◦ ) mean the design face
attached to the octree is vertical.
Direction of The direction calculated from the normal vector Influence of orientation with
the design associated with the octree block (coloured by degree). respect to stress and rock mass
surface. properties.
Characterise how OB and UB are
affected by face orientation.

Distribution of the direction for each octree along the design

surface (dots in the image). The direction varies from 0 to 360◦ .
Effective Calculated by measuring, at regular radial angle Local stability conditions –
radius intervals, the Distance from a point located on the deformation of excavated span.
factor selected surface of the stope to the edge of that surface. Characterise how OB and UB are
(ERF). A maximum value is obtained towards the center of the affected by the size and geometry of
surface. the faces.

Distribution of the ERF for each octree along the design surface

(continued on next page)

7
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Table 3 (continued )
Parameter Concept Interpretation Example

(dots in the image). Warm colours mean the octrees are further
away from the surface edges and have higher stability
concerns.
Undercut. The stope design as well as the drifts are ‘wrapped’ in a Local stress conditions.
convex hull. Loss of confinement.
Distance (coloured by meter) is found from an octree Characterise how OB and UB are
block to the nearest point on the convex hull surface. affected by the complexity of the
geometry and the cutting of the
stope by the drives.
In comparison with the stope
geometry complexity parameter,
the undercut considers the full
geometry of the void and the
impact the drives have.

Distribution of the undercut for each octree along the design

surface (dots in the image). Warm colours indicate that, due to
the presence of the drives and/or the shape of the stope, the
octrees are further away from the simplified shape that wraps
the stope and drives, and highlights areas where there can be a
loss of confinement due to how the drives cut the stope.
Distance to Distance to the nearest drift in regard to the octree’s Deconfinement and gravity effect
drives. position. on design surface areas near the
structures.
Damage induce during
development.
Characterise how OB and UB are
affected by the presence of drives.

Distribution of the Distance to the drives for each octree along

the design surface (dots in the image). Cold colours (0 m)
means the octrees are on the drive or close and warm colours
indicates the octrees are further away from the drives.
RQD. Value from the nearest block of the geotechnical block Characterisation of how OB and UB
model mapped to the octree. are affected by the rock mass
The block model is built using borehole data. The nearest quality.
RQD value is mapped to the block with the nearest
Distance varying between 0 and 30 m with 20 % within 5
m, 60 % within 10 m and 90 % within 15 m.
The RQD is directly logged from the core.

Distribution of the RQD for each octree along the design surface
(dots in the image). Warm colours will highlight areas where
there is poor rock mass quality.

represents the average predicted value from all the trees used in the model. (revenue in regard to a cut-off grade). The stope performance assessment
The standard error is calculated from these trees using the infinitesimal is detailed as follows.
jackknife approach32 by omitting observations used and recalculating the
average predicted value. This process is repeated, and the variance in the 3.3.1. Stability assessment
average predicted value is calculated. The standard error is then used to The resolution of the predictions enables the assessment of the areas
calculate the probability density function (PDF) and the cumulative den­ where OB and UB are expected, and using the standard error calculated for
sity function (CDF) for the predicted octree, allowing to determine a each predicted octree, the probability of exceeding a given tolerated
prediction interval within which the projected distance is expected. threshold of OB or UB. The tolerance for OB and UB will vary from one
Different geometries can be outputted using the different percentiles of the mine site to another as it depends on the geological and mining environ­
prediction interval. The probability of observing a specific projected dis­ ment, dictating how the OB and UB impact the profitability and stability of
tance can be determined using the PDF, and the probabilities of exceeding the stope and subsequent stopes. The predicted OB and UB are assessed by
a given projected distance can be determined using the CDF. first plotting the octrees in a 3D view. The octrees are coloured using
markers, showing the predicted Distance or the probability of exceeding a
3.3. Stope performance assessment given threshold of OB or UB. This allows to determine the areas of concern
and the probability for critical OB or UB. The percentage of the design
Once the predictions are obtained, the predicted projected distances surface that is possibly exceeding a given threshold is calculated and
are analysed to understand where and to what probability OB and UB analysed by plotting a cumulative chart of the proportion of octrees
will occur in the stope along with the associated economic impact exceeding the threshold at each percentile of the CDF.

8
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Fig. 9. 3D view of stope A; a) The design geometry with the drives; b) The octree data structure coloured per face.

The CDF is also used to analyse the prediction interval for the final The calculation of the grade value at an octree resolution also means
mined geometry. For example, three different geometries (20th, 50th that a cumulative distribution of the NSR value for each tonne in the
and 80th percentile of the CDF) are generated by calculating the pro­ stope or volume type is calculated (Fig. 7a). As a footnote, the total
jected Distance of the octrees for these percentile, giving a prediction tonnage for every category is indicated to enable relative importance of
interval (Fig. 5a). Since it’s a cumulative function, each percentile rep­ the cumulative volume. This chart is used to compare the distribution of
resents a probability of observing a projected distance smaller or equal the grade, or in this case the NSR, in each volume (design, mined, OB
to the Distance calculated for that percentile. This is presented for one and UB) and determine if the predicted geometry follows the design NSR
octree in Fig. 5b. At the 20th percentile mark, the projected Distance for tonne distribution. It is also used to calculate the number of tonnes
an octree is 1.2 m. There is a 20 % chance the actual Distance will be under the cut-off grade for each volume type. The economic value of the
smaller and a 80 % chance it will be larger. For the example in Fig. 5b, OB and UB is also assessed, evaluating the proportion of waste and ore to
there is a 50 % probability (prediction interval between 20 % and 80 %) determine how the OB and UB impacts the stope dilution and loss of ore.
that the projected Distance for the given octree will be between 1.2 m These tonnes are also queried using a selection tool on the graph and
and 2.8 m. The 50th percentile represents the geometry generated by the plotted in 3D view to identify their spatial location. Fig. 7b is an example
model (mean case) and is used to determine where the OB and UB will of a selection of OB octrees plotted in 3D and coloured using their NSR
most likely occur. To evaluate worst case scenarios for OB, percentiles value. Octrees with an NSR value bellow cut-off are semitransparent to
over 50 % are analysed. The same is done inversely for UB by looking at highlight the areas where dilution is expected to occur.
probabilities under 50 %. The different geometries generated using the The dilution and loss of ore are calculated using the cut-off grade and
probabilities are then reconciled with the stope design to calculate the are used to quantify the profitability of the stope. There are different
predicted OB and UB volumes and perform an economic assessment. ways of calculating the dilution based on the literature and from what is
seen on mine sites. The measures implemented are listed in Table 2.
3.3.2. Economic assessment These values are compared between the different predicted geometries
The NSR designates the value of the ore after milling and smelting, and to assess the impact of OB and UB on dilution.
is often used as the economic parameter for designing the stope as it en­
compasses all the contained metals. Therefore, the NSR value is used for
the present analysis. The block model data is spatially associated to the 3.4. Root cause assessment
octree data structure to calculate the grade value in the OB, UB and other
volumes. The weighted average NSR value as well as the total metal in the Once the prediction for the stope is generated and the economic
octree block are calculated from the grade blocks that intersect the octree information is compiled, different geometries are compared to assess the
block (Fig. 6). If backfill information is available, the grade value for the expected performance. If the predicted stope performance is satisfactory
octree which contains fill is adjusted. This information is assessed by (economically and geometrically) and the probabilities for critical OB or
plotting the data in a 3D view in order to identify the areas of low value or UB are within the mine site’s risk tolerance, engineers can proceed with
high value and compared with the probabilities of OB and UB. Since each the final planning of the stope. If concerns about the stability and
octree is associated to one or more volume type (OB, UB, design, mined profitability are flagged during the stope performance assessment, a root
and recovered) the average and total NSR is also summed up per face and cause assessment is done to identify the factors influencing the predicted
per stope for the different volume types. This information is calculated for dilution or loss of ore from the OB and UB and determine if possible
the different probabilities to assess and compare the economic impact of modifications to the design are practically viable.
the OB and UB on the profitability of the stope. The areas of concern in the stope are highlighted through the stope
performance assessment. The critical parameters for each area are

9
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Fig. 10. The three predicted geometries surfaces and per octree (20th, 50th and 80th percentiles of the prediction interval).

10
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Table 4 plotting the octrees in a 3D view to visualise the area of concern and the
Predicted per stope summary for the geometries 20th, 50th and 80th percentiles parameter value for that area. The variable importance of the model is
of the prediction interval. also used to direct this assessment as it identifies the general importance
20 % 50 % 80 % Mined of each parameter for predicting using the model.
Designed volume (m3) 32 212 32 212
Predicted volume (m3) 24 680 30 451 37 295 29 626 4. Case study
Volume difference (%) − 23 − 5 16 − 8
OB (m3) 316 1706 5995 1817 Prominent Hill is an underground copper deposit with gold and silver
UB (m3) 7847 3467 913 4400
located in South Australia. The mining method is transversal stopes to
Recovered (m3) 24 365 28 745 31 299 27 812
Compliance to design (%) 75 85 82 82 extract the ore using a primary–secondary sequence. Stope dimensions
vary depending on the sector but are typically 20 m wide by 20 m long or
30 m by 30 m and up to 90 m high. The orebody reaches a depth of
analysed to identify the primary causes of OB or UB by plotting the approximately 1 km.
critical parameter versus the projected Distance in bivariate charts. The
octrees from the previously mine stopes are also plotted to compare how
the current octrees correlate to OB or UB. The analysis is facilitated by

Fig. 11. 3D view of the probability of exceeding a 0 m or 2 m threshold and a cumulative distribution chart of the percentage of octrees above the threshold value for
the probability geometry (CDF percentile).

11
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Table 5 Table 7
Dilution and loss of ore summary for the three different geometries of the pre­ Per face summary of the NSR values for the footwall and hanging wall OB and UB
diction interval and the mined geometry. for the predictions and mined geometry. Profit is calculated according to
Equation (5).
Measure 20 50 80 Mined Equation
% % % 20th 50th 80th Mined
percentile percentile percentile
Planned dilution (%) 18 18 Wastedesign
∗ 100
Oredesign OB Average NSR 44 46 45 49
Predicted dilution (%) 22 30 33 19 Wastepredicted footwall ($/tonne)
∗ 100
Orepredicted Total ($) 45 435 215 672 479 536 50 629
Unplanned dilution (%) 1 3 8 1 WasteOB Profit ($) − 6196 − 18 296 − 52 927 − 1033
∗ 100
Orepredicted OB Average NSR 97 77 74 79
Loss of ore in UB (%) 25 11 3 13 OreUB hanging ($/tonne)
∗ 100
Oredesign wall Total ($) 1081 45 373 348 666 165
Predicted recovered ore 73 86 104 92 Orepredicted 490
∗ 100
(%) Oredesign Profit ($) 521 16 013 111 800 60 749
UB Average NSR 86 88 87 85
footwall ($/tonne)
Total ($) 323 502 119 098 33 690 344
Table 6 233
Profit ($) − 134 320 − 51 057 − 14 238 − 141
Per stope summary of the NSR values per volume type for the predictions and
743
mined geometry. Profit is calculated according to Equation (5). UB Average NSR 85 83 81 77
20th 50th 80th Mined hanging ($/tonne)
percentile percentile percentile wall Total ($) 550 279 178 723 44 117 127
529
Design Average NSR 123.1 Profit ($) − 224 670 − 70 446 − 16 904 − 44
($/tonne) 718
Total ($) 14 182 410
Profit ($) 8 420 969
Predicted Average NSR 127 121 116 125
fully captured. Furthermore, blasting is not an exact cutting tool
($/tonne)
Total ($) 11 199 266 13 245 603 15 504 464 13 327 compared to a tunnel-boring machine, implying some uncertainty
633 around the position of the final surface. The variability and uncertainty
Profit ($) 6 776 073 7 790 844 8 828 682 7 996 in the parameters is considered to a certain degree by incorporating a
580 probabilistic approach.
OB Average NSR 52 63 77 103
($/tonne)
The list of parameters presented in Table 3 is based on the root-cause
Total ($) 61 391 393 416 1 653 413 671 analysis and the data made available by the mine site. No geomechanical
663 parameters are included in this analysis, as no data other than an RQD
Profit ($) 2633 83 108 576 743 345 block model was available. Producing a geotechnical model at a 1 m
613
scale resolution for data currently available at underground mining sites
UB Average NSR 109 108 102 99
($/tonne) is challenging. If such data are incorporated into the analysis, great care
Total ($) 3 044 536 1 330 223 331 359 1 511 must be taken to ensure they are compatible with the other data. There
315 are also currently no stress parameters that have been developed at an
Profit ($) − 1 647 − 713 233 − 169 031 − 748 octree resolution.
529 025

4.2. Random forest model overview

4.1. Data overview
For this case study, only the transversal stopes will be used. Of the 55
The stope database is comprised of 73 stopes mined between 2018 stopes, four stopes were excluded from the model due to quality con­
and 2022 inclusively (Fig. 8). Reconciliation on a per stope, per face and cerns (missing drill rings and stope tunnelling into adjacent backfill). A
per octree basis was conducted for each stope. Octrees associated to first model was built to evaluate the prediction performance of the
drives are excluded as they represent existing voids, and backfill faces random forest for this dataset. 75 % of the stopes were used to build the
are excluded as well, due to their different strength and stress condi­ model and 25 % for evaluating the model’s performance. The model has
tions. This is to ensure the quality and consistency of the analysis. Of the an accuracy of 66 % (proportion of octrees correctly classified) for
73 stopes, 55 stopes are transversal stopes and 18 are longitudinal classifying an octree as OB or UB and, using the mean predicted case, has
stopes. a projected distance error smaller than half a meter 33 % of the time,
Economic, geological and operational data was made available by smaller than a meter 58 % of the time and smaller than 2 m 83 % of the
the mine site. The approach for selecting the critical parameters is time. This statistical performance matches what was observed by14 at
presented in.34 The stopes, as well as a selection of critical geotechnical, other mining sites and is. Therefore, it is determined that random forest
geometrical and operational parameters (Table 3) were used to build a can be used for this dataset. A final model was built from the 40 stopes in
random forest model. A stope is used for each example, displaying the the database that were mined before the stope used for this case study.
octree parameter values according to the accompanying legend. Expla­ Octrees (drives, muck pile, backfill and floor) were excluded to ensure
nations in this table discuss the underlying influences that these pa­ the quality of the model.
rameters aim to characterise. These parameters require the CMS, stope
design geometry, drift surveys, drill rings and structural model to be 5. Results
available for each stope. It is assumed and accepted that there is
different levels of inherent uncertainty in the data arising from the Stope A is used as an example to demonstrate the stope performance
complexity of the underground geological and mining environment and root cause assessment (Fig. 9a). The octree data structure was built
where the variation in the geomechanical properties of the rock mass, as at a 1 m resolution and six faces were delimitated (Fig. 9b). The critical
well as the variation between the mined and planned data, cannot be parameters presented in Table 3 are computed for each octree. Octrees
representing drives, as well as the floor, were filtered out for the

12
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Fig. 12. NSR value per octree, per volume and per tonne for the three different geometries (20th, 50th and 80th percentiles of the prediction interval). A cut-off value
of 50 $/tonne is set. Octrees below the cut-off value are semitransparent. The mined volume in the cumulative charts represents the predicted mined volume. The
black line indicates the cut-off limit.

13
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Table 8 5.1.3. Probabilistic analysis

Parameters ordered by variable importance calculated from the The probabilities are also used to determine the chance of exceeding
random forest model. a given threshold. For example, in Fig. 11, a 0 m (separation of OB and
UB) and 2 m OB threshold are used. If we look at the area around the
drives in the hanging wall (circled in black), there is a 60 %–70 % chance
of observing OB and a 20 % chance that it will exceed 2 m. A cumulative
distribution chart can be used to determine the percentage of octrees
(equivalent to the portion of the surface) below or above the threshold
for each probabilistic scenario. For example, the portion of the surface
that would be overbroken (below the curve) for the mean case (proba­
bility of 50) is 33 %, meaning there is a 50 % chance of observing 33 %
or more OB for the surface. This surface area drops to 6 % for OB over 2
m for the same probability.

5.2. Economic analysis

Following the same process as the stability analysis, the economic

analysis is first done on a per stope or per face basis to assess the global
economic performance of the stope. With the geological block model, an
NSR value is attributed to each octree and, for this example, a fictitious
but plausible cut-off grade of 50 $/tonne is used.

5.2.1. Per stope analysis

predictions. These octrees will be attributed a prediction of 0 m after the The dilution and loss of ore summary for the predictions and mined
predictions are generated to maintain the point density across the sur­ geometry of stope A is presented in Table 5 The geometry that best
face for building the predicted mined surface. matches the design and the mined geometry of the three is the median
case with the main difference being the predicted dilution (30 %
compared to a mined dilution of 19 %). The predicted dilution is 12 %
5.1. Stability analysis higher than the planned dilution with only 11 % loss of ore and a pre­
dicted recovery of ore of 86 %. The predicted dilution interval varies
The predictions were generated for stope A octrees and surfaces were between 22 % and 33 % (over-predicting the actual dilution) with the
generated for three different geometries based on the CDF percentile: unplanned dilution varying between 1 % and 8 %. The loss of ore varies
the 20th, 50th and 80th percentile (Fig. 10). The 20 % and 80 % cases between 25 % and 3 %, and the expected ore recovery is expected to
were chosen arbitrarily for this example and could vary for a mine site vary between 73 % and 104 %, encompassing the amount actually
based on their preferences. mined.
The impact of the dilution and loss of ore on stope profit can also be
5.1.1. Per stope analysis computed as the NSR value is available. The NSR value per stope and per
The predicted surface built from these cases is first used to determine face are shown in Table 6 and Table 7. Profit is calculated according to
the global stability of the stope by quantifying OB and UB on a per stope Equation (5).
or per face basis (Table 4). The volume difference, recovered volume and
compliance to design are calculated according to Equations (2)–(4). The Profit = (Average NSR per tonne – cut − off) ∗ Tonnes 5
volume difference and compliance to design are based on.5,6 For com­ If mining goes according to the design, the stope should generate a
parison, the mined results obtained from the CMS are also included. The profit of 8 420 969$ with an average NSR of 123.1 $/tonne. The model
prediction interval gives a prediction volume with a volume difference predicts a profit between 6 776 073$ and 8 828 682$ with an average
with the design between − 23 % and 16 %. For the predictions, the OB NSR between 127 and 116 $/tonne according to the 20th and 80th
could vary between 316 and 5995 m3 and the UB could vary between percentile with the mean predicted profit being 7 790 844$ at 121
7847 and 913 m3. Overall the prediction interval encompasses the $/tonne. The loss of profit can be associated to the dilution in the
mined volumes with the 50th percentile being the scenario that matches footwall where the average NSR is around 44–46 $/tonne, which is
best the mined volume with a 3 % difference. below the 50$ cut-off, and the loss of ore in the UB which could vary
(predicted − design) between 169 031$ and 1 647 529$. However, UB in the footwall and
Volume difference = ∗ 100 2 hanging wall only represent 18 %–20 % of the loss of profit, indicating
design
that the majority could be recuperated with the adjacent stopes. Extra
Recovered = Design − UB 3 ore can be recuperated in the hanging wall as the average NSR in its OB
varies between 74 and 97 $/tonne.
Design − UB After mining, the stope generated a profit of 7 996 580$ at 125
Compliance to design = ∗ 100 4
Design + OB $/tonne which closely matches the 50th percentile. 748 025$ is lost in
the UB with 25 % coming from the footwall and hanging wall, reducing
5.1.2. Per octree analysis the amount of ore that can be recuperated from the adjacent stopes. The
The stope stability can then be assessed at a finer resolution, looking OB and UB in the footwall is closer to the 20th percentile with dilution in
at the projected Distance per octree (Fig. 10). Analysing the results: the footwall as predicted.

• The mean case predicts OB for the majority of the hanging wall and 5.2.2. Per octree analysis
footwall with UB occurring more in the corners and on the left side of With the stope economic summary considered, the analysis is refined
the sidewalls. on an octree basis to identify where the costs and profits are located on
• Potential large OB (over 3 m) is seen in the footwall along the middle. the stope. The quantification of the NSR per octree also allows to
• The side walls present probable large UB (over 3 m). quantify the distribution of the NSR value per tonne for the design,

14
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Fig. 13. Multiple 3D view of the design octrees of stope A coloured with the different critical parameters.

mined, OB and UB volumes (Fig. 12). The NSR distribution for the 20th of the top of the hanging wall. Loss of ore can be a concern for the rest of
and 50th percentile closely follows the design. The 80th percentile has a the stope, although the higher-grade ore is mostly located in the side
higher proportion of tonnes below 100 $/tonne than the design. The UB walls. This ore could be recuperated when mining the adjacent stopes for
follows a similar distribution for the three percentiles with 90 %–95 % the side walls. Looking at the other volumes and the different geome­
being over the cut-off limit but has less value than the tonnes in the tries, there is an 80 % probability to generate dilution in the footwall
predicted geometry. Waste in the OB varies between 40 % and 65 %. The (see OB for the 20th percentile) and leaving ore in the corners of the
total tonnes of OB represent, however, only a fraction of the predicted faces (see UB for the 80th percentile).
mined tonnes. This information is also assessed by analysing the octrees
associated with the different volumes (design, predicted, OB and UB) for
5.3. Root cause analysis
each predicted geometry and coloured by NSR value as shown in Fig. 12.
Octrees with values under 50 $/tonne are semitransparent. Looking at
Given the predictions of the stope, the site’s risk and OB or UB
the design, dilution concerns are in the left side of the footwall and part
tolerance, and the site’s economic goals (no ore left behind or minimise

15
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Fig. 14. Original and alternate blast design proposed for optimisation analysis.

dilution, for example), the engineer can determine if the potential footwall. Looking at Fig. 13, the following observation can be made:
outcome of the stope does not meet the objectives of the stope, the
respect of the design geometry and the profitability of the stope. If the • The footwall presents high energy (Fig. 13d) at the top and bottom of
outcome is judged to be not meeting the design requirements, a root the face and a relaxation zone due to undercut (Fig. 13j).
cause analysis can be done to determine the key parameters controlling • There is also a lower RQD area (50–70 %, Fig. 13k) at the bottom of
the predicted performances. For stope A, if the goal is to minimise the face.
dilution, dilution in the footwall would be the main concern as the
average NSR value is below the cut-off value. If the loss of ore is the main Potential optimisation could involve changing the charging against
concern, UB in the other faces could result in potentially millions of the footwall and/or potentially some ground support – although, in this
dollars’ worth of ore being left behind while OB would imply limited case, the impact of ground support has not been analysed. The undercut
dilution and some ore recovery. When considering the economics in issue would remain.
isolation, it would be reasonable to aim to minimise UB and tolerate If the loss of ore were the main concern, UB would need to be
some additional OB. Additional factors considered by the engineer may minimised in the other faces. Looking at Fig. 13, the following obser­
also impact to what degree this trade-off is made; for example, ore vations can be made:
fragmentation and the impact on the stability of adjacent stopes to be
mined in the future. • UB is more pronounced in the corners (Fig. 13a).
• UB in the corners is mostly associated to the drilling pattern (Fig. 13c
5.3.1. Interpretation of the model and d).
The first step in the root cause analysis, as described in Section 3.4, is
to consider the importance of the variable calculated by the model. Potential optimisation could be achieved by changing the drilling
When building the model, a value is attributed by the model to each pattern.
parameter based on its importance for predicting stope performance. This root cause analysis only considered the parameters used in the
The order of parameter importance is presented in Table 8. From this model. Other factors could also have an impact on stope performance
order, it is interpreted that performance is determine by separating the and should not be excluded.
different faces (dip and orientation) as the side walls, hanging wall,
footwall and crown perform differently. After which, the position of the 5.4. Exploring other designs
drives (undercut) and the distribution of the blasting energy will impact
where the OB and UB will occur. Going down the list, the size of the The model structure enables to modify certain design parameters and
faces, the angle of the boreholes to the face, the gravity effect caused by rerun the model in order to explore how stope performance could
the intersection of the drives and the rock mass quality will decrease in fluctuate to work towards an optimal design that matches the mine
impact. operator’s goals. For this case study, the design geometry or the drill
rings could be altered as they are considered in the model and certain
5.3.2. Visual analysis aspects could be modified in the long-term or short-term design. Stope A
Given the general interpretation of stope performance by the model, is used as an example again. Given the disseminated ore in the hanging
the analysis is extended, looking at the stope in the 3D view and col­ wall and the possibility to recuperate the ore in the corners with the
ouring the octrees with the different parameters (Fig. 13) to determine, adjacent stopes, the goal for the exercise will be to minimise dilution in
for specific areas, which parameter is critical. As determined before, if the footwall. Using the information from the root cause analysis and the
dilution were the main concern, one would want to limit OB in the model, blast rings against the footwall will be omitted in order to

16
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Fig. 15. Original and alternate projected distance and economic predictions for optimisation analysis.

evaluate how the footwall performance could change. Fig. 14 shows the The predicted dilution interval drops by 9 % from 22 % to 33 %–13 % to
original blast design and the alternate blast design as well as the vari­ 24 %. The average NSR in the footwall UB drops from 86 to 88 $/tonne
ation of the blasting energy proxy for the footwall, which diminishes. to 77 to 81 $/tonne, indicating more waste would be left behind.
The alternate predictions are presented in Fig. 15. The probability of However, the predicted UB increases by up to 1500 m3 from the original
OB over 2 m diminished for the majority with UB becoming more scenario, and the loss of ore interval increases by 2 %.
probable. Waste can be seen to increase in the UB while decreasing in the This alternate economic performance that implies omitting the drill
OB. ring close to the footwall would reduce the dilution overall while
Table 9 provides a comparison of the overall and footwall predicted maintaining the economic performance of the stope, all while mining
performance for the original and alternate design. The profit intervals between 6 and 8 % less rock than the original predicted scenario. This
are similar for the two scenarios, with the alternate design predicting a would reduce the operation time dedicated to the stope and fill material.
slightly smaller profit for the 20th and 50th percentile. However, the Overall, the economic risk from dilution would be minimised as the
average NSR increases for the alternate case, reducing the interval from predicted dilution interval drops and the average NSR per tonne
11 $/tonne (127–116 $/tonne for the original scenario) to 9$/tonne increases.
(130–121 $/tonne for the alternate scenario). This optimisation exercise was fast to complete (approximately 15
The predicted OB volume is also reduced from a possible 6000 m3 to min for one stope) and showed that limiting the energy proxy in the
4000 m3, increasing the predicted average NSR in the footwall OB from footwall would help reduce the dilution and cost of waste processing
below the cut-off (44–46 $/tonne) to over the cut-off (53–60$/tonne). from the footwall. Caution needs to be taken when optimising through

17
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

Table 9 of which at a meter-scale resolution allowing to move away from

Comparison of the predicted performance between the original and alternate averaged values. While only a single case study is presented in this
design. article, this methodology is meant to be flexible and applicable to
Design 20th 50th 80th various mine sites. The type of deposit and the critical factors impacting
percentile percentile percentile the mine site stope performance can affect the methodology’s
Total value ($) Original 11 199 266 13 245 603 15 504 464 performance.
Alternate 10 702 314 12 763 733 15 061 394 The knowledge obtained from the predicted stope performance
Profit $ Original 6 776 073 7 790 844 8 828 682 assessment can be used by the engineer to assess the expected perfor­
Alternate 6 560 348 7 687 677 8 812 804
mance of the stope and design, and consequently provide paths to
Predicted Average Original 127 121 116
NSR ($/tonne) Alternate 130 126 121 optimise the stope. Possible optimised scenarios can be generated using
Predicted recovered Original 73 86 104 the model due to the flexibility of the model to incorporate any pa­
ore (%) Alternate 72 88 105 rameters such as the operational parameters calculated (drill and blast).
Predicted dilution (%) Original 22 30 33 Possible optimisation will, however, depend on the critical factors
Alternate 13 19 24
OB (m3) Original 316 1706 5995
identified through the root cause analysis and if they are used in the
Alternate 32 498 3982 model. Current optimisation through the model can remain limited, as
Footwall avg NSR Original 44 46 45 some critical parameters could not be considered in the models at this
($/tonne) in OB Alternate 53 60 54 stage but will improve with time as more parameters are quantified on
Loss of ore in UB (%) Original 25 11 3
an octree basis, thus improving the model at the same time. This
Alternate 28 13 4
UB (m3) Original 7847 3467 913 approach enables one to rerun the predictions after modifying certain
Alternate 9355 4374 1286 parameters to evaluate how it would impact the expected mined
Footwall avg NSR Original 86 88 87 geometry.
($/tonne) in UB Alternate 81 77 80
Compliance to design Original 75 85 82
(%) Alternate 71 85 85
7. Conclusion
Volume difference (%) Original − 23 − 5 16
Alternate − 29 − 12 8 This article presented, through a case study, a new stope design
approach that uses geometrical, geotechnical and operational parame­
ters, as well as reconciliation data, measured at a meter-scale resolution,
this approach as it is an empirical approach and not analytical. There­
to predict a stope’s OB and UB location and magnitude on the design.
fore, proposing a new design that does not resemble the ones used in the
Through the multivariate statistical model (random forest), a standard
model would mislead the engineer in their assessment of the predicted
error was calculated for each prediction, allowing to generate different
stope performance. A good understanding of how the model works and
possible geometries of the mined geometry and determine the proba­
the mine site environment is needed. However, it remains that this
bility of large OB and UB based on the mine site’s tolerance. The eco­
proposed approach offers a new innovative and powerful tool for pre­
nomic performance was assessed by mapping the grade block model to
dicting stope performance at the design stage, and a pathway for facil­
the octree data structure in order to determine the NSR value, the
itating optimisation at the design stage.
dilution and the loss of ore for the stope. This assessment was then used
to determine the root causes of the OB and UB based on the parameters
6. Discussion and summary
used in the model which can be used by the engineer to determine the
need and possible ways to optimise the design.
The proposed methodology for predicting stope performance using
This new methodology works towards a full integration of an eco­
multivariate statistics (random forest) and data at an octree resolution
nomic and stability analysis from predictions at the design stage by
enabled to obtain a probabilistic prediction of a stope’s expected ge­
enabling to extract more information and precision from the predictions,
ometry. The calculation of the standard error of each prediction has
and build an understanding of the controlling factors of OB and UB and
allowed to evaluate the possibility of large OB or UB occurring (pre­
how they impact the profitability of a stope. This knowledge is built on
diction interval) and, given the resolution of the predictions, to know
the site’s data and enables to integrate operational data to facilitate the
where it would occur. Contrary to a single prediction, this approach
optimisation of stope design and improve performance.
allows to determine the probability of OB or UB exceeding an acceptable
This is a significant leap forward in open stope design technology
threshold and allows the engineer to integrate the economic impact in a
when compared to the qualitative stability output of the stability graph
stope performance analysis.
or the very crude estimation of ELOS and ELLO.
The octree data structure enables to incorporate the grade block
model for assessing the economic impact of the OB and UB. The inclu­
Statements and declarations
sion of the economic assessment also allows to evaluate the impact the
OB and UB will have on dilution and loss of ore and, using the different
The research leading to these results received industry funding from
geometries, how the increase and decrease in OB and UB impact stope
BHP, Olympic Dam mine and Prominent Hill mine; Glencore Mount Isa
profitability. The economic assessment only considers the NSR value and
Mines; IAMGOLD Corporation, Westwood mine; MMG Limited, Dugald
the cut-off limit. Therefore, only dilution, loss of ore, total value and
River mine and Newmont, Tanami mine.
average value can be determined. The impact of mucking time, frag­
mentation and how the waste mixes with the ore is not considered in this
CRediT authorship contribution statement
analysis.
Existing methods for predicting stope performance only gives a
Benoît McFadyen: Writing – original draft, Methodology, Formal
general stability assessment (stable, unstable or caving) or an ELOS or
analysis. Martin Grenon: Writing – review & editing, Conceptualiza­
ELLO value for each face depending on the method used. The predictions
tion. Kyle Woodward: Writing – review & editing. Yves Potvin:
don’t allow to know where the OB and UB would occur in the face. They
Writing – review & editing.
also limit the economic assessment. This proposed methodology pre­
sents a big step forward in how the predictive tools will impact and
Declaration of competing interest
improve future stope performance. It provides a unique path towards
integrating economic and stability assessment of stope performance – all
The authors declare that they have no known competing financial

18
B. McFadyen et al. International Journal of Rock Mechanics and Mining Sciences 178 (2024) 105778

interests or personal relationships that could have appeared to influence 14. McFadyen B, Grenon M, Woodward K, Potvin Y. Predicting open stope performance
at an octree resolution using multivariate models. J South Afr Inst Min Metall. 2023;
the work reported in this paper.
123(6):309–320.
15. Suorineni FT, Kaiser PK, Tannant DD. Likelihood statistic for interpretation of the
Data availability stability graph for open stope design. Int J Rock Mech Min. 2001;38:735–744.
16. Diederichs MS, Kaiser PK. Rock instability and risk analyses in open stope mine
design. Can Geotech J. 1996;33(3):431–439.
The authors do not have permission to share data. 17. Idris MA, Nordlund E. Probabilistic-based stope design methodology for complex ore
body with rock mass property variability. J Min Sci. 2019;55:743–750.
Acknowledgement 18. Lane K. The Economic Definition of Ore: Cut-Off Grades in Theory and Practice.
Brisbane, Au: COMET Strategy Pty Ltd; 1988.
19. Wang H, Webber T. Practical semiautomatic stope design and cutoff grade
The authors thank the sponsor for permission to use their data in this calculation method. Min Eng. 2012;64(10):85.
paper. This research would not be possible without the six industry 20. Pakalnis RT. Empirical Stope Design at the Ruttan Mine, Sherritt Gordon Mine LTD.
Dissertation: University of British Columbia; 1986.
sponsors of this project and the ongoing collaboration from the mine 21. Villaescusa E. Geotechnical Design for Sublevel Open Stoping. Boca Raton, FL: Taylor &
site’s personnel. The sponsors were BHP, Olympic Dam Mine and Francis; 2014.
Prominent Hill mine; Glencore Mount Isa Mines; IAMGOLD Corporation, 22. Potvin Y, Wesseloo J. Stope design. Geotechnical Design for Underground Metalliferous
Mines Course. Perth, Au, Australian Centre for Geomechanics; 2022.
Westwood Mine; MMG Limited, Dugald River Mine and Newmont, 23. Urli V, Esmaieli K. A stability-economic model for an open stope to prevent dilution
Tanami mine. We gratefully acknowledge both the corporate and indi­ using the ore-skin design. Int J Rock Mech Min. 2016;82:71–82.
vidual support. 24. McFadyen B, Grenon M, Woodward K, Potvin Y. Assessing stope performance using
georeferenced octrees and multivariate analysis. Min Technol. 2023;132(2):132–152.
https://doi.org/10.1080/25726668.2023.2205080.
References 25. Harris PH, Wesseloo J. mXrap Software, Version 5, Australian Centre for Geomechanics.
The University of Western Australia; 2015. https://mxrap.com/.
1. Potvin Y, Hadjigeorgiou J. The stability graph method for open-stope design. In: 26. Woodward K, Harris P. Stope reconciliation and analysis. mXrap Software App.
Hustrulid WA, Bullock RL, eds. Underground Mining Methods, Engineering Australian Centre for Geomechanics. The University of Western Australia; 2019.
Fundamentals and International Case Studies. Colorado. Society for mining, 27. Breiman L. Random forest. Mach Learn. 2001;45:5–32. https://doi.org/10.1023/A:
metallurgy, and exploration; 2001:513–520. Inc. 1010933404324.
2. Mathews KE, Hoek E, Wyllie DC, Stewart SBV. Prediction of Stable Excavation Spans 28. Wright MN, Ziegler A. Ranger: a fast implementation of random forests for high
for Mining at Depth below 1,000 M in Hard Rock Mines. Ottawa, Ca: CANMET; 1981: dimensional data in C++ and R. J Stat Software. 2017;77(1):1–17. https://doi.org/
Report 802–1571. 10.18637/jss.v077.i01.
3. Potvin Y, Woodward K, McFadyen B, Thin I, Grant D. Benchmarking of stope design 29. R Core Team. R: A Language and Environment for Statistical Computing. R Foundation
and reconciliation practices. In: Wesseloo J, ed. Proceedings Of the Second for Statistical Computing; 2021. URL HYPERLINK https://www.R-project.org/.
International Conference on Underground Mining Technology. 2020. https://doi.org/ https://www.r-project.org/.
10.36487/ACG_repo/2035_14. Perth, AU. 30. Kazhdan M, Bolitho M, Hoppe H. Poisson surface reconstruction. Proceeding of
4. Guido S, Grenon M. Contributions to geomechanical stope optimization at the goldcorp Eurographics Symposium on Geometry Processing. 2006 (Sardinia, Italy).
eleonore mine using statistical analysis. Proceeding of the ISRM International 31. Girardeau-Montaut D. CloudCompare software, version 2; 2022. Retrieved from
Symposium—10th Asian Rock Mechanics Symposium. 2018 (Singapore). http://www.cloudcompare.org.
5. McFadyen B. Détermination de Facteurs Contrôlant la Stabilité des Chantiers Souterrains 32. Wager S, Hastie T, Efron B. Confidence intervals for random forests: the jackknife
Pour une Mine Sujette à la Sismicité. Université Laval; 2020. Dissertation https and the infinitesimal jackknife. J Mach Learn Res. 2014;27.
://corpus.ulaval.ca/entities/publication/5b4c3a9b-9179-4821-ae3b-3939c84c233 33. McFadyen B, Woodward K, Potvin Y. Open stope design; beyond the stability graph.
e. AUSIMM. Proceeding of Underground Operators Conference 2021. 2021 (Perth, AU).
6. Woodward K, McFadyen B, Potvin Y, Wesseloo J. Probabilistic Stope Design. MRIWA 34. Potvin Y, Grant D, Mungur G, Wesseloo J, Kim Y. Practical stope reconciliation in
Project No. M489. Perth, Au: The Australian Centre for Geomechanics; 2019. large-scale operations Part 2, olympic Dam, South Australia. Proceeding Of Seventh
7. Miller F, Potvin Y, Jacob D. Laser measurement of open stope dilution. Cim Bull. International Conference & Exhibition on Mass Mining. Sydney, AU: MassMin; 2016.
1992;85:96–102. 35. Potvin Y. Empirical open stope design in Canada. Mining and Mineral Process
8. Woolmer D, Jones E, Taylor J, Baylis C, Kewe D. Use of drone based lidar technology Engineering. PhD Thesis. University of British Columbia; 1988. https://open.library.
at olympic Dam mine and initial technical applications. In: Castro R, Báez F, ubc.ca/soa/cIRcle/collections/ubctheses/831/items/1.0081130.
Suzuki K, eds. Proceedings of the Eighth International Conference & Exhibition on Mass 36. Nickson SD. Cable Support Guidelines for Underground Hard Rock Mine Operations.
Mining. 2020. https://doi.org/10.36487/ACG_repo/2063_38. Santiago, Chile. Dissertation: University of British Columbia; 1992. https://open.library.ubc.ca/med
9. Clark L. Minimizing Dilution in Open Stope Mining with a Focus on Stope Design and ia/stream/pdf/831/1.0081080/2.
Narrow Vein Longhole Blasting. Dissertation: The University of British Columbia; 37. Guido S, Grenon M, Germain P. Stope performance assessment at the Goldcorp Eleonore
1998. https://open.library.ubc.ca/soa/cIRcle/collections/ubctheses/831/items/1. mine using bivariate analysis. Proceeding of AfriRock 2017 Rock Mechanics for Africa.
0081111. 2017 (Cape Town, SA).
10. Wang J. Influence of stress, undercutting, blasting and time on open stope stability and 38. Henning JG, Mitri HS. Numerical modelling of ore dilution in blasthole stoping. Int J
dilution. Dissertation. University of Saskatchewan; 2004. https://harvest.usask.ca/h Rock Mech Min. 2007;44:692–703.
andle/10388/etd-11032004-094152. 39. Shnorhokian S, Mitri HS, Moreau-Verlaan L. Stability assessment of stope sequence
11. Hughes R. Factors Influencing Overbreak in Narrow Vein Longitudinal Retreat Mining. scenarios in a diminishing ore pillar. Int J Rock Mech Min. 2015;74:103–118.
Dissertation: McGill University; 2011. 40. Suorineni FT. Effects of Faults and Stress on Open Stope Design. Dissertation:
12. Qi C, Fourie A, Du X, Tang X. Prediction of open stope hangingwall stability using University of Waterloo; 1998. https://uwspace.uwaterloo.ca/handle/10012/342.
random forests. Nat Hazards. 2018;92:1179–1197. https://doi.org/10.1007/ 41. Suorineni FT, Kaiser PK, Tannant DD. Determination of fault-related sloughage in
s11069-018-3246-7. open stopes. Int J Rock Mech Min. 1999;36:891–906.
13. Adoko AC, Saadaari F, Mireku-Gyimah D, Imashev A. A feasibility study on the 42. Scoble MJ, Moss A. Dilution in underground bulk mining: implications for
implementation of neural network classifiers for open stope design. Geotech Geol production management. Geol Soc London, Spec Publ. 1994;79(1):95–108.
Eng. 2022;40:677–696. https://doi.org/10.1007/s10706-021-01915-8.

19