Professional Documents
Culture Documents
net/publication/277022455
CITATIONS READS
2 6,835
2 authors, including:
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Fidelis Tawiah Suorineni on 14 November 2017.
ABSTRACT
With decreasing profit margins, optimising the size of stopes to minimise dilution is a step towards
achieving a productive and profitable mining operation. The stability graph was developed
to provide guidance in the determination of stope sizes to control dilution in bulk mining.
Unfortunately, this graph is qualitative and stopes can only be described as stable, unstable or
cave. The alternative to the qualitative stability graph is the equivalent linear overbreak slough
(ELOS) stability graph. The original stability graph is only applicable to wide orebodies while
the ELOS stability graph only applies to narrow-vein orebodies and does not provide explicit
quantitative dilution values. The objective of this research is to develop a generalised quantitative
dilution-based stability graph independent of orebody width. To achieve this objective a total of
226 stope case studies on open stope performances were gathered from six current underground
metalliferous mining operations located across Australia. The data was statistically analysed using
Logistic regression and the Bayesian likelihood discrimination method to produce quantitative
dilution-based stability graphs independent of orebody widths. The graphs provide the mining
engineer the flexibility to design open stope sizes based on what dilution amounts are acceptable
to a given operation. While the graphs were developed from a statistically significant database,
additional data could improve the confidence and reliability in their use. It should also be noted
that while the database is from Australian mines, the graphs could be used elsewhere with similar
geological and mining conditions to those used in the study.
INTRODUCTION
In the global competitive mineral commodities market there the conventional stability graph developed by Mathews et
are high pressures on mining operations to maximise reserves, al (1981) and the equivalent linear overbreak slough (ELOS)
optimise production and increase profits by reducing costs. stability graph (Clark and Pakalnis, 1997) are only applicable
The conventional stability graph by Mathews et al (1981) is to wide and narrow-vein orebodies respectively.
a qualitative design tool used to determine the potential
stability of stopes. When used correctly it can provide mine BACKGROUND
planners with optimal stope dimensions. Optimal stope sizes The stability graph method for open stope design was
result in less development costs and in turn, lower mining developed by Mathews and co-workers at Golder Associates
costs and dilution. (Mathews et al 1981) for predicting stable spans in open stope
The paper presents a generalised quantitative dilution- mining at depths below 1000 m.
based stability graph independent of orebody width for The stability graph is a plot of the stability number N
open stope design. This is achieved by applying Logistic against a hydraulic radius HR/shape factor S (Figure 1). In
regression and Bayesian likelihood discriminant methods Figure 1, if the stability state of a stope plots in the stable zone
to stope performance data collected from various operating it means that surface has a high probability of being stable,
underground metalliferous mines in Australia. A generalised if a surface plots in the unstable zone it implies that surface
quantitative dilution-based stability graph has more attached has a high probability of suffering some degree of failure
value to miners and mine planners as they appreciate dilution which could be avoided or improved with support. The
numbers more and their implication to the profitability of the unstable zone is sometimes referred to as the supportable
operation than merely knowing whether a stope is stable, zone. A stope surface plotting in the cave zone implies there
unstable or cave in the case of the conventional stability is a high probability about 30 per cent of the stope surface
graph by Mathews et al (1981). As the proposed stability will slough but not in the sense of caving as in block caving.
graph is orebody width independent it can easily be applied The three stope stability states in Figure 1 are separated by
effectively to both wide and narrow-vein orebodies, whereas what are referred to as transition zones.
1. SAusIMM, School of Mining Engineering, UNSW Australia, Kensington NSW 2052. Email: alexander.papaioanou@gmail.com
2. MAusIMM, School of Mining Engineering, UNSW Australia, Kensington NSW 2052. Email: f.suorineni@unsw.edu.au
rather than quantitative cavity monitoring survey (CMS) •• When availability of site data allowed, other dilution
data. CMS is used to monitor stability of cavities, and is equations were used to cross reference dilution values
frequently used in assessing open stope performance for determined from ELOS. Comparisons were made between
dilution determination. Also, the inclusion of case studies of calculated dilutions and site provided stope dilutions for
caving from underground block caving, longwall coal mining validation.
and cut-and-fill is inappropriate (Suorineni, 2010) as this •• Dilution was determined via a comparison between the
results in mixing entry and non-entry mining methods with a initial designed stope shape and the final resultant stope
consequence on safety. Some amount of failure is acceptable shape measured by CMS. This method is the most objective
in non-entry mining methods but cannot be tolerated in entry way of determining stope dilution and overbreak. By
methods such as cu-and-fill. comparing the initial design size/volume and the final
The original stability graph was designed for large bulk design size/volume, and using Geovia Surpac™ to
‘non-entry’ stopes. Recently, the stability graph method has determine the difference in volumes and size a dilution
been inappropriately applied to the design of open stopes value could be determined and validated. Figures 2 and 3
in narrow vein orebodies resulting in misleading stability illustrate this method.
outcomes, further hindering the reputation of the stability
graph method (Suorineni, 2010).
DILUTION
While there are numerous definitions for dilution, this project
primarily focuses on unplanned overbreak, which occurs as
a result of instability within a stope. The Equivalent Linear
Overbreak Slough, ELOS (Clark and Pakalnis, 1997), is an
indirect quantitative measure of dilution. ELOS is defined in
Equation 5. The ELOS stability graph database is developed
from narrow vein mines.
While the ELOS concept can be applied to narrow and wide
orebodies when it is converted into per cent dilution in stopes
for the two types of orebodies the difference in values and
potential impact on profitability is huge. Equation 6 provides
a means for converting ELOS values into per cent dilution
and forms the basis for this paper.
Volume of slough
ELOS = (5)
Stope wall surface area
METHODOLOGY
In order to develop a generalised quantitative dilution-based
stability graph independent of orebody width for open stope FIG 2 – Stope design and cavity monitoring survey outline.
design, a new stope stability graph database is required.
Due to the sensitive nature of the stope stability data the overlap between data categories from which the transition
names of the mines have been kept confidential by identifying boundaries between the data groups can be inferred.
them by letters only. The equiprobability-contour method also known as the
approximate method is used to determine the degree of
Statistical approaches overlap between data categories from which the transition
In order to increase confidence and reduce subjectivity and boundaries between the data groups can be inferred.
bias, statistical analysis was used to define the boundaries Discriminant analysis is characterised by its ability to
between dilution categories. Two statistical approaches were successfully determine the degree of overlap in data
used: Logistic regression and Bayesian likelihood statistic. categories. An increasing degree of overlap can result in
Logistic regression is used to analyse and statistically diminishing success and usefulness of discrimination.
delineate dilution zone boundaries within the stability Consequently, a separability index, ϱ, was developed by
graph. Logistic regression analysis has advantages over other Suorineni, Tannant and Kaiser (2001) in order to account for
traditional regression techniques such as ordinary linear the amount of overlap between categories. This separability
regression, which does not effectively consider the discrete index, ϱ, is defined as the ratio of the distance between the
nature of the dependent variable being examined (Laio, centroids of two dilution categories such as dilution zone
1994). Logistic regression applies a non-linear transform categories X (<5 per cent) and X < (5 - 10%) and the pool
to transform a linear combination of independent variables variance of both of the two data categories S as in Equation 9.
to a binary output of zero or one. For each of the resultant
outcomes of case variables, a logit function was used to ϱ = X 5 - 10% - X < 5% (9)
estimate the probability of the event occurring. S
The logit model (Equation 7) was adopted from Mawdesley By applying the equiprobability-contour method to the data,
(2002). Data input into the logit regression model composed contours of equiprobability can be determined surrounding
of two independent variables: the modified stability number, the centroid of a category, where the overlap between the
N’, and the hydraulic radius, HR. Then, using a binary logit equiprobability contours of different categories, define the
function the probability of a stope surface fitting into the boundaries between the categories.
two dependent variables is assessed. This predicted logit
The likelihood ratio, Λ, defined in Equation 8 represents the
probability value can be compared with the original values,
likelihood function of two data categories or normal density
allowing for analysis of misclassified cases. In binary logistic
functions of a bivariate system in a pair of data categories which
regression the logit value produced represents the natural
form a multivariate data set (Suorineni, 1998). For dilution-
logarithm of the odds, in which the odds indicate the relative
based stability graph analysis, the pair of data categories
probability of the case being classified into one of the two
is the two dilution categories for case stope surfaces. Each
categories (Mawdesley, 2002).
category is assessed with its following category to determine
The Logit model is given by Equation 7, which incorporates the degree of overlap and to develop the equiprobability
both N’ and HR for use within the stope stability graph. contours and resulting boundaries. The likelihood ratio Λ
The values α, β1 and β2 are estimated using the maximum defined in Equation 8 can be represented by Equations 10 and
likelihood method which is derived from binomial distribution simplified in Equation 11. Equation 11 defines the transition
(Bergerud, 1996). boundaries between different categories. Coefficients β0 and
β1 can be determined from simultaneous equations of two
z = a + b1lnHR + b 2 lnN l (7) data groups (Suorineni, 1998).
Microsoft Excel™ add-on software XLSTAT™ was used to b0 log10 HR + b1 logN l + k = 0 (10)
determine the Logit values and predicted Logit probability
values.
HR = 10 b0 + b1 log10 N l (11)
Another statistical method for determining the dilution
category boundaries in the stability graph is the Bayesian
likelihood statistic discriminant analysis using equiprobability
contours and the likelihood ratio approach. This approach
RESULTS AND DISCUSSION
was used by Suorineni (1998). The likelihood ratio approach
in discriminant analysis is derived from Bayes’ Discrimination
Results
rule which states that ‘assign the object to the group with the After the data validation, all data was composited within a
highest conditional probability’. When applied to the stability Microsoft Excel™ database where it was preprocessed to
graph it considers all the posteriori probabilities of a specific identify any outliers. In total 226 case histories were used
case point falling into one of two discrete categories based on from six operating Australian underground metalliferous
the highest probability of the dependent variable and how open stope mines.
other case points have been assigned. The likelihood ratio (Λ) After the data validation, all data was composited within
and equiprobability methods assume that two data groups a Microsoft Excel™ database where it was preprocessed to
are normally distributed (Suorineni, 1998). The likelihood identify any outliers. In total 226 case histories were used
ratio is given by Equation 8, where the example of the two from six operating Australian underground metalliferous
dilution categories <5 per cent and 5–10 per cent is used. open stope mines.
In order to subject the 226 case study stope surfaces to
f(5 - 10%) (X)
the logistic regression procedure Microsoft Excel™ add-
K= (8)
f< 5%(X) on software XLSTAT™ was used to perform the logistic
regression. Three parameters where entered into XLSTAT™
The equiprobability contour method also known as the as variables: dilution category, (the dependent variable) the
approximate method is used to determine the degree of modified stability number and hydraulic radii. The binary
logit model utilised limited data from the two categories Using Equation 12 the predicted logit probability value p
at a time. In order to develop the transition boundary lines from Figure 4 can be transformed into a predicted log odd’s
between dilution categories the upper and lower dilution value z and regression coefficients α, β1 and β2 in Equation 6
categories were compared together in the binary logit can be written as Equation 13 which represents the boundary
model. As such <5 per cent category was compared with line between dilution categories.
5–10 per cent category and 5–10 per cent category compared
p
with 10–15 per cent category and so on. This process was Z = ln c m (12)
repeated to create four distinctive logit data sets. 1-p
Logistic regression is used to determine the location of
the transition boundaries which can be used to demarcate (z - a - b1ln (HR))
different dilution categories. In order to determine the Nl = e b2 (13)
position of boundaries cumulative logit values were
determined for each dilution category pair (<5 per cent and Repeating the process for each paired dilution categories
5–10 per cent, etc). To achieve the cumulative distributions and applying Equation 13, the generalised dilution-based
each dilution category is plotted along with the inverse stability graph independent of orebody width for open stope
of each respective category on a cumulative distribution design was developed with the logistic regression (Figure 5).
graph as shown in Figure 4. The crossover point between Subjecting the 226 case study stope surfaces to the Bayesian
the upper bound dilution category such as <5 per cent discriminant analysis is similar to the logistic regression
cumulative distribution function intersects the inverse approach. The statistical software SYSTAT™ developed
lower point cumulative distribution function representing by Sigmaplot was utilised to analyse the data using
the logit probability value which defines the separation line discriminant analysis and to represent dilution categories
(Mawdesley, 2002). The intersecting Logit value determines with equiprobability contours. Again, three parameters
the vertical axis intercept for the dilution category boundary were entered into SYSTAT™ as variables: dilution category
while, the inclination of the transition boundary is (the dependent variable), the modified stability number and
represented by regression coefficients α, β1 and β2. hydraulic radii. Similar to binary logistic regression two
dilution categories were analysed at a time. The resultant
graph of equiprobability contours formed the basis for the
boundary lines between dilution category pairs.
Using Equation 11, the boundary line between two dilution
category pairs can be determined. The boundary line
corresponds with the intersection between corresponding
equiprobability contours as shown in Figure 6. Figure 6
represents the <5 per cent dilution and 5–10 per cent dilution
category pairs. Each equiprobability contour determined by
SYSTAT™ surrounds the centroid of the dilution category
with outwards increasing probability. Probability ranges in
Figure 7 are ten per cent, 20 per cent, 40 per cent, 60 per cent
and 80 per cent. The point A in Figure 6 is the point where
FIG 4 – Cumulative distribution function <5 per cent and 5–10 per cent. the stope surface has 40 per cent chance of having a dilution
FIG 5 – Generalised dilution-based stability graph independent of orebody width for open stope design developed with logistic regression.
FIG 6 – Bayesian likelihood discrimination <5 per cent to 5–10 per cent dilution equiprobability-contour plot.
between five and ten per cent and a 60 per cent chance of it does not correct for misclassification errors. Furthermore,
having a dilution <5 per cent. when the data categories are not equal to each other due to
The resultant equiprobability contours between dilution unequal priori probabilities, the discriminant is bias towards
categories is shown in Figure 7 that are used to define the the larger data group which is the >5 per cent and 5–10 per cent
boundaries between dilution categories (Figure 8). Figure 8 dilution categories in this study. The logistic regression model
shows the stability graph developed with the Bayesian used to develop boundary lines is limited by its fixed gradients,
likelihood discrimination and the respective dilution values. whereas the Bayesian likelihood discrimination model used
The ability to statistically develop boundary lines based upon can account for varying gradient of boundary lines, there by
the likelihood method and equiprobability contours results in accounting for data inequalities in the data groups.
reduced subjectivity. Bayesian likelihood discrimination can be
extended to estimate predictive errors in the stability graphs. CONCLUSIONS AND RECOMMENDATIONS
The project involved a data collection campaign between May
Discussion and July 2014. In total, 226 case histories were collected from
Logistic regression analysis is capable of defining isoprobability six current operating underground metalliferous open stope
contours representing boundaries between sets of data but mine sites located across Australia.
FIG 7 – Generalised dilution-based stability graph independent of orebody width for open stope design
developed with Bayesian likelihood discrimination and equiprobability-contour plot.
FIG 8 – Generalised dilution-based stability graph independent of orebody width for open stope design
developed with Bayesian likelihood discrimination with dilution point values (per cent).
The collected data was preprocessed to identify and Clark, L and Pakalnis, R, 1997. An empirical design approach for
eliminate outliers. The resultant data was used to develop estimating unplanned dilution from open stope hanging walls
a quantitative dilution-based stability graph independent and footwalls, in Proceedings CIM AGM, Calgary (Canadian
Institute of Mining, Metallurgy and Petroleum: Westmount).
of orebody width using two methods of statistical
analysis, Logistic regression and the Bayesian Likelihood Diederichs, M S and Kaiser, P K, 1996. Rock instability and risk
discrimination method. analyses in open stope mine design, Can Geotech J, 33:431–438.
A quantitative dilution-based stability graph is more Laio, T F, 1994. Interpreting probability models: logit, probit, and
meaningful to a miner than a qualitative or average overbreak other generalised linear models, in Sage University Paper series
on Quantitative Applications in the Social Sciences, 12:88–100 (Sage:
depth because miners clearly understand the implications of
Thousand Oaks).
dilution numbers for their mines. The graph also gives mine
planners and designers the flexibility to design stope sizes Mathews, K E, Hoek, E, Wylle, D C and Stewart, S B V, 1981. Prediction
of Stable Excavation Spans for Mining at Depths below 1000 meters in
based on their acceptable dilution levels. The quantitative
Hard Rock, pp 36–110 (Golder Associates: Vancouver).
dilution-based stability graph is also orebody width-
independent. An orebody width independent stability graph Mawdesley, C A, 2002. Predicting rock mass cavability in block
has the benefit of not being misapplied. The conventional and caving mines, PhD thesis (unpublished), Julius Kruttschnitt
Mineral Research Centre, University of Queensland, Brisbane.
ELOS stability graphs which are orebody-width dependent are
currently being misused as it is not obvious to users that these Mawdesley, C A, Trueman, R and Whiten, W, 2001. Extending the
graphs are orebody-width (narrow versus wide) dependent. Mathews stability graph for Open-stope design, Transactions
The use of statistical methods to define transition boundaries of the Institutions of Mining and Metallurgy, Mining Technology,
110(1):A27–A39.
between dilution categories eliminates subjectivity and bias.
Mitchell, R J, Olsen, R S and Smith, J D, 1982. Model studies on
It is recommended that additional data be collected and
cemented tailings used in mine backfill, Canadian Geotechnical
the dilution boundaries fine-tuned to improve the predictive Journal, 19:14–28.
reliability of the graphs. It is also recommended to continue
Nickson, S D, 1992. Cable support guidelines for underground hard
the research using the Bayesian likelihood discriminant
rock mine operations, Master thesis (unpublished), University of
method because of its capacity to account for data category British Columbia, Vancouver.
in-balances.
Potvin, Y, 1988. Empirical open stope design in Canada, PhD thesis
(unpublished), University of British Columbia, Vancouver.
ACKNOWLEDGEMENTS
Stewart, S B V and Forsyth, W W, 1995. The Mathews method for
The authors would like to thank the numerous industry open stope design, The CIM Bulletin, 88(1036):45–83.
contacts for their valuable support and contributions
Suorineni, F T, 1998. Effects of faults and stress on open stope design,
throughout the project. The project was funded by the School PhD thesis, University of Waterloo (unpublished), Waterloo.
of Mining Engineering Research Grant (SRG).
Suorineni, F T, 2010. The stability graph after three decades in use:
experiences and the way forward, International Journal of Mining,
REFERENCES Reclamation and Environment, 24(4):307–339.
Barton, N, Lien, R and Lunde, J, 1974. Engineering classification of Suorineni, F T, Tannant, D D and Kaiser, P K, 2001. Likelihood
rock masses for the design of tunnel support, Rock Mechanics, statistic for interpretation of the stability graph for open stope
6(3):192–210. design, International Journal Rock Mechanics and Mineral Science,
Bergerud, W A, 1996. Introduction to Logistic regression models with 36(7):891–906.
worked forestry example, Ministry of Forests Research Program,
in Biometrics Information Handbook, 7(26):146–148.