2 Implementation of an expert system approach for risk assessment.

Safety in Mines Research Advisory Committee
Final Project Report
Technology transfer of the risk assessment
tool for rock-related risk issues
A.v.Z Brink
Research Agency:
CSIR: Division of Mining Technology
Project Number:
SIM 03 02 01
Date:
March 2005
1
Executive summary
The outcome of earlier SIMRAC projects, GAP 608 and Gap 714 was a prototype software
package that was a combination of an expert system approach for determining overall risk
assessment, a risk control recommendation, and a basic Graphic Information System (GIS)
approach for graphically overlaying input parameters and the output risk. The expert system
used a Bayesian probabilistic approach to combine various types of workplace-related
information, a quantified seismological environment, and exposure of underground staff to
provide an overall rock engineering (RE) risk rating. The risk assessment was dynamic in that it
was recalculated each time any input parameters changed, for example each time a seismic
event was recorded in the area of interest.
The objective of this project was to provide clear documentation of an actual application of a
rock engineering risk assessment tool as developed earlier. As the validity of risk assessment
cannot be shown by relating to specific short term incidents, the project attempted to evaluate
the method after an extensive period of use with a large data set. This could not be achieved in
real time because of severe limitations in access to real-time seismic data on mines, as well as
system interfacing to an operational mine’s database for rock engineering parameters.
The project ran a case study demonstration over two years on, primarily, seismological data.
The continuous risk assessment on dynamically changing input parameters was proven. The
case study allowed for correlation between assessed risk and the subsequent hazards
experienced.
The case study included a continuous assessment of RE risk on a mining section on the
Ventersdorp Contact Reef (VCR) in the Far West Rand. Over a time window of one year, a
significant majority of the larger events (Magnitude > 1) in the immediate vicinity of the operating
panels were preceded by a noteworthy increase in the risk levels. Similarly, a significant
minority of all high risk levels did not manifest into a relatively large event. These conclusions
were subjective and reached by viewing the risk display over a period of time. An objective and
quantified measure of determining success rate was not proposed in this project. Absolute
quantification of success rate should be possible, but then again, the implication is that this
concept is being presented as a seismic prediction tool, rather than RE risk assessment.
The project may be criticised in that the non-seismic aspects of RE risk assessment did not
contribute to the case study evaluation. The interfacing problems to a mine’s RE database did
not allow for the inclusion of real non-seismic data and the researcher had to resort to simulated
data. However, the concept of using continuously changing input evidence whilst determining
the probability of an output hypothesis being true, was well proven.
The software platform used was appropriate for development and proof of concept, but an
operational tool, interfacing to mine systems and user interfacing, still needs to be developed.
The project output is compiled on a CD, comprising all the evaluation coding, input data and
results. Future software implementation of the now proven concept should be undertaken by a
professional software developer, without needing to understand the basic principles employed
in this demonstration/evaluation version. The remaining outstanding aspects of this potentially
useful tool are mainly the real-time data and user interfacing. It is recommended that these
aspects be addressed in a future project.
2
Acknowledgements
The author would like to express his gratitude to the Mine Health and Safety Council for
financial support of project GAP 03 02 01 and their patience. Malcolm Drummond from
GeoVision is acknowledged for his contribution in developing the visualisation platform for this
project.
Dr Steve Spottiswoode are recognised for his valuable comments and probing questions.
3
Table of contents
Executive summary ......................................................................................................... 2
Table of contents ............................................................................................................. 4
List of figures ................................................................................................................... 5
1
Summary of related earlier SIMRAC work (GAP 608 and GAP 714) ................. 6
1.1
Introduction .................................................................................................................. 6
1.2
Development of a software tool for the integrated assessment of rock engineering risk
.................................................................................................................................... 7
1.3
Development of probabilistic risk assessment using an expert system philosophy ....... 8
2
Implementation of an expert system approach for risk assessment. .................. 9
2.1
Defining the input parameters for RE risk assessment ............................................... 10
2.1.1 The prior probability .......................................................................................... 10
2.1.2 The individual probability parameters................................................................ 11
2.1.3 Temporal change in event rate as an input parameter towards RE risk
assessment ...................................................................................................... 11
2.1.4 Apparent Volume as an risk indicator of possible rock burst damage................ 12
2.1.5 Energy Index as an risk indicator of possible rock burst damage ...................... 13
2.1.6 Peak ground motion as an risk indicator of possible rock burst damage ........... 14
3
Evaluation of the risk assessment software. .................................................... 16
3.1
Incorporating rock mechanics parameters ................................................................. 16
3.2
A case study on continuous risk assessment ............................................................. 17
4
Conclusions ..................................................................................................... 18
5
References ....................................................................................................... 19
Appendix A ...................................................................................................................... 1
Appendix B .................................................................................................................... 31
4
List of figures
Figure 1.1 Production losses as result of seismicity as a function of the magnitude of the
damaging seismic events .................................................................................... 7
Figure 1.3
A schematic of the integrated approach towards risk assessment .......................... 8
Figure 2.1 Rate of change in the number of events per time unit, with the red contour relating to
a doubling in the event rate. .............................................................................. 12
Figure 2.2 Rate of change in the Apparent Volume as observed in a short-term long-term ratio
.......................................................................................................................... 13
Figure 2.3 As in Figure 2.1 and Figure 2.2 but this time the rate of change in Energy Index is
overlain on the mining operation........................................................................ 14
Figure 2.4
A plot of the number of times during the previous 30 days that an estimated peak
particle velocity of .2 m/s was exceeded ............................................................ 15
Figure 3.1 Total working area to be considered in terms of risk to the underground personnel.
The area includes the active panels and the back areas ................................... 16
Figure 3.2 The process of linking a geographical area of an active panel to a set of attributes 17
5
1
1.1
Summary of related
(GAP 608 and GAP 714)
earlier
SIMRAC
work
Introduction
The primary output of this project is to provide clear documentation of actual applications of a
rock engineering (RE) risk assessment tool on gold and platinum mines. In order to do this,
reference is needed to the earlier work that resulted in the concept and development of a RE
risk assessment tool to be evaluated. GAP 608 (Brink et al., 2000) reviewed all applicable work
in South Africa that was carried out with the aim to define various methods to ‘quantify’ hazard
and/or risk in the context of the rock engineering stability of the rock mass. GAP 714 (Brink et
al., 2002) developed a ‘beta’ version of a software package employing an expert system
approach in combining these various hazards and risk parameters.
A number of SIMRAC projects have addressed the issue of hazard and/or risk in gold and
platinum mines, with a specific emphasis on the hazard and/or risk posed by rock-related
incidents. Often, the distinctions between hazard and risk were unclear. It advisable to review
the respective definitions of hazard and risk from Mine Health and Safety Act No 29 of 1996:
Hazard is a physical situation, object or condition, which, under specific circumstances
has the potential to cause harm, whereas risk is a measure of the likelihood that some
specific harm, arising from an incident (a particular hazard).
The generic equation:
Risk = Hazard * Vulnerability
( 1-1)
Was extended by Menoni et al.,1997, to:
Risk = Seismic Hazard * Induced physical hazard * Systemic Vulnerability
( 1-2)
where Induced physical hazard = rock related damage in working place, with the potential to
cause injuries, and triggered by ground motion; support failure; or fall of ground. (definition
extended by Brink et al., 2000)
Systemic Vulnerability = Exposure of people; economic vulnerability; quality of information.
(definition extended by Brink et al., 2000).
The first term, Seismic Hazard, was subsequently also extended to included probability of
occurrence. For this reason it could be argued that it is rather Seismic Risk.
The final risk formula developed for software implementation could therefore be written as:
RE risk = Seismic Risk * Systemic Vulnerability
( 1-3)
In this case Seismic Risk is the probability that a specific working place can experience damage
as result of a seismic event within a specified time period (that is Seismic Hazard and Induced
physical hazard as in Equation (1-2)). The Systemic Vulnerability is a combination of numerous
weighting factors.
RE risk is a unitless parameter that now quantifies the rock related risk in a working place at a
specific time. The risk may be specific as the risk of injury and/or fatality or the economic risk
(loss of equipment or production) or both. Being a unitless parameter, RE risk has a
6
disadvantage in that the comparison of different areas on a mine, or the same area at different
times, is not possible if exactly similar parameters are not used as input.
The deliverable of GAP 714 was a prototype software package that is a combination between
an expert system approach for determining overall risk assessment, a risk control
recommendation, and a basic GIS approach for graphically overlaying input parameters and the
output risk. The expert system uses a Bayesian probabilistic approach to combine various types
of workplace-related information, a quantified seismological environment, and exposure of
underground staff to provide an overall RE risk rating. The risk assessment is dynamic, in that it
is recalculated each time any input parameter changes - for example, every time a new seismic
event is recorded in the area of interest.
RE risk assessment should, however, not be seen as an alternative form of rock burst
prediction.
The authors of the report on GAP 714 (Brink et al., 2002) had the opportunity to evaluate the
various risk assessment methodologies and tools during their consultation with the mining
industry. Figure 1.1 shows the distribution of production shifts lost relative to the size of the
damaging seismic events. It cannot, of course, be concluded that Mag 1,5 to 2,0 events are a
greater hazard than a Mag 3,5 event but that, with their shorter reccurrence time, the former, in
fact, pose a greater risk. It was clear that the RE risk, and even seismic risk does not
correspond exclusively to the few very large seismic events.
Production risk
% of shifts lost
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0.50
1.00
1.50
2.00
2.50
3.00
3.50
4.00
4.50
Magnitude
Figure 1.1 Production losses as result of seismicity as a function of the magnitude of
the damaging seismic events (from Lenhardt, 1998)
1.2
Development of a software tool for the integrated
assessment of rock engineering risk
An objective of the earlier work was to develop computer code that would integrate various
seismic and non-seismic hazard and/or risk parameters in a single risk assessment. A further
objective was to facilitate this integration through an expert system approach. Figure 1.2
schematically demonstrates the approach used in the development of the code for RE risk
assessment. The non-seismic (or general rock mechanics) data tends to be linked to face
names rather than geographic positions. Similarly, the seismic data is purely geographical, with
no links to the actual face names.
7
RE
database
+ face
files
Expert
system
evidence
probabilities
Input matrix
100x100x100
Expert system
evaluation of input
evidence and display
of output risk
hypothesis
Seismic
database
Output display array
100x100
Figure 1.2 A schematic of the integrated approach towards risk assessment
Basically, the mining configuration is overlain with a 1 km x 1 km grid with a grid resolution of
10 m x 10 m. Each grid element has its own attributes describing the various static risk
conditions (typically the conditions describing the vulnerability of the excavation) and the
dynamic conditions (parameters describing the seismicity as experienced at that specific grid
element). Each grid element allows for 100 attributes. The software, therefore, has to manage
an overall risk matrix with dimensions of 100 x 100 x 100.
Visualisation of input and output parameters is accomplished through extensions to a basic 2D
data-integration environment developed by GeoVision. The environment offers the ability to
dynamically view several spatially related datasets together in a single window. Special
pipelines were also developed by GeoVision for the import of mine plans, creation and/or
editing of static input datasets (input rasters), and the dynamic import of spreadsheet data
(output raster).
1.3
Development of probabilistic risk assessment using an
expert system philosophy
The Bayes' Theorem was adopted as an expert system philosophy for RE risk assessment. The
basic concept of this philosophy is that a hypothesis should be developed that is tested against
some evidence. In this case the hypothesis states that a rock related incident (rock fall or rock
burst damage) would occur within the next eight hours. A prior probability, P(H), is set. This is
the probability that such a rock-related incident may occur during the next eight hours (the
probability of the hypothesis being true) without any associated evidence.
Evidence associated with the hypothesis being true, is for example, the observation that a high
event rate is prevailing. The probability of observing such evidence, that is a high event rate,
whilst the hypothesis is also true, p(E|H) is determined from previously observed information.
Similarly, the probability of observing the evidence, but with a hypothesis that is not true,
p(E|~H), is set.
Based on Bayes' Theorem, the probability of observing the hypothesis H if the evidence E has
been observed is the probability p(H|E).
p( H | E ) 
p( E | H ) p( H )
p( E | H ) p( H )  p( E |~ H ) p(~ H )
( 1-4)
8
(p(H) is the prior probability of the hypothesis being true and p(~H) the prior probability of the
hypothesis not being true.)
Similarly, the probability of observing the hypothesis H if the evidence E has not been observed,
p(H|~E):
p( H |~ E ) 
p(~ E | H ) p( H )
p(~ E | H ) p( H )  p(~ E |~ H ) p(~ H )
( 1-5)
The initial importance (or Rule Value) of each item of evidence is given as:
RV 
p( H | E)  p( H |~ E)
( 1-6)
A typical expert system will have a number of possible hypotheses and even more items of
evidence that can be associated with these hypotheses. For the purpose of testing the
hypothesis of a large rockfall and/or burst within the next eight hours, the expert system only
considers this single hypothesis, tested against the available items of evidence.
The main advantage of employing an expert system philosophy is that it can cater for
uncertainties. The expert system philosophy, from XMaster®, (Chris Naylor Research Limited,
Naylor, 1998 and Naylor, 2000) allows for the actual response to whether specific evidence is
observed, to be between a 'yes' and a 'no'.
It is possible to calculate a new posterior probability for each hypothesis, conditional upon the
response.
p( H | R)  p( H | E ) p( E | R)  P( H |~ E ) p(~ E | R)
( 1-7)
With a user response of 'yes' then p(E|R) would be 1 because there would be certainty that the
evidence had been identified. Similarly, with a user response of 'no' then p(E|R) = 0 if there
were certainty that the evidence had not been identified.
A 'do not know' will result in p(E|R) =0.5, with all other situations resolved by interpolation.
The next step is to calculate the final posterior probability of the hypothesis being true if various
bits of evidence of various degrees of confidence are observed. The simplest approach is to
calculate the final posterior probability by the successive application of the appropriate formulae
for a single item of evidence to each of the items of evidence in turn.
Initially, the hypothesis starts off with the given prior probability. This prior probability is updated
by the application of one item of evidence, which may or may not be observed with a given
degree of uncertainty, to produce a new posterior probability. This new posterior probability is
then used as the new prior probability for a re-application of the same equation to the next item
of evidence. The process continues in this way until all of the items of evidence have been
accounted for.
2
Implementation of an expert system approach for
risk assessment.
The following parameters are combined through a probabilistic approach:
The potential for significant ground motion in a specific excavation,
the intensity of such ground motion,
the vulnerability of the excavation, and
the exposure and quality of data used as weighting factors.
9
To achieve this any parameter that can be shown to be relevant evidence that a specific
hypothesis may be true, should be considered. The hypothesis in this case is that a rock related
incident with the potential to cause injuries and/or fatalities might occur within the following shift.
The code was developed to allow for up to ten seismic parameters to constitute the seismic risk.
These parameters may be used as evidence to quantify the probability for a large ground
motion at a particular point on or close to the reef plane. The respective position of the seismic
events is given in 3-D but the risk evaluation is done on a 2-D plane of the mine workings.
The seismic input parameters used are the respective changes in short term / long term ratio
as observed in event rate, apparent volume and energy index. In adopting the Bayes’ Theorem
expert system approach, the outcome that is hypothesised, is that a rock related incident might
occur during the current or next shift within the boundaries of the grid resolution, i.e. 10 m.
Smoothing is applied to include the input parameters of adjacent grid elements, which results in
an effective resolution of about the dimensions of a typical panel length.
2.1
Defining the input parameters for RE risk assessment
With reference to equation ( 1-4);
p( H | E ) 
p( E | H ) p( H )
p( E | H ) p( H )  p( E |~ H ) p(~ H )
,
the various input parameters need to be defined.
2.1.1 The prior probability
The first input to the risk assessment process, is the prior probability, p(H). That is the
probability that the hypothesised outcome will occur without any evidence being observed that
either supports or rejects the hypothesis. p(H) is estimated by looking at all the related incidents
in area of interest and over a period of time. The prior probability is then:
p( H ) 
# of incidents
(# of grid pnts)  (# of shifts)
( 2-1)
where # of incidents is the number of rock related incidents that caused physical harm;
# of grid pnts is the number of grid elements included in the working areas; and
# of shifts is the number of shifts during the review period.
The test area selected was a mini-longwall on the VCR in the Far West rand comprising seven
panels. The area was subdivided into a 1000 m by 1000 m area with a 10 m grid resolution.
Seismicity for a period of one year was analysed (975 events with Mmin = - 0,5 and MMax = 3.)
The number of events per grid point larger than Magnitude 1, or even the number of such
events that may influence a grid point, was too low to be able to quantify any changes in
parameter behaviour. For the purpose of quantifying the prior probability, it was decided to
group all the seismicity in this area.
A total of 70 events larger than Magnitude 1 occurred in this area. The prior probability of an
event happening during any eight hour shift is therefore the number of events/the number of
shifts per year.
The result for this area under consideration is a prior probability, p(H) = 0.064.
10
2.1.2 The individual probability parameters
Each input parameter or piece of evidence is quantified in terms of the probability of observing
the evidence with the hypothesis being true, p(E|H) , and the probability of observing the
evidence, but with a hypothesis that is not true, p(E|~H). Equation ( 1-7) allows for the actual
response to whether specific evidence is observed, being between a 'yes' and a 'no'.
The specific individual parameters used were:

Temporal change in event rate;

Temporal changes in Apparent Volume;

Temporal changes in Energy Index ;

Number of PPV (peak particle velocity) exceeding a specific threshold.
Each parameter is described and quantified in the following sections.
2.1.3 Temporal change in event rate as an input parameter towards
RE risk assessment
The concept of using a change in event rate was developed by Brink et al., (2001) for the
application towards early warning of goafing at Moonee Colliery in Australia. The same concept
was applied in this project. A short term event rate (for the previous ten events), versus a longer
term event rate (for the previous fifty events) was used. Should a short term versus longer ratio
of larger than two be observed, it was considered as a 100% positive indication that this specific
input evidence was observed. The opposite would be a ratio of 0,5 implying a 0% indication
that a change in event rate was observed.
As before for prior probability, P(H), the same area and seismicity is used. The probability of
observing the evidence (a high event rate ratio) with a true hypothesis (an event larger that
Mag. 1) during the following shift is p(E|H) = 0,67. This means that 67% of the larger events
were preceded by an event rate ratio of at least two. Similarly, the probability of experiencing a
shift (eight hours) with the evidence being true, but with the hypothesis not being true, p(E|~H),
was 0,016. In other words, during 1,6% of the total number of eight hours shifts the event rate
parameter has exceeded the predetermined value, but no event larger than a magnitude 1
occurred during that, or the following shift.
The Bayes Theorem input parameters for event rate is summarised in Table 2-1.
Table 2-1 The input evidence as quantified for event rate.
Parameter
p(E|H)
p(E|~H)
Ryes
Rno
Event Rate
0,67
0,016
2,0
0,5
Should only event rate be used as input evidence, Figure 2.1 depicts the area of interest with
the red contour value highlighting grid points that at the time of the ‘snap shot’ experienced an
event rate ratio of more that two.
11
Figure 2.1 Rate of change in the number of events per time unit, with the red contour
relating to a doubling in the event rate.
2.1.4 Apparent Volume as an risk indicator of possible rock burst
damage
Apparent Volume (m3) relates to the volume of rock with coseismic inelastic strain and is
calculated as: (From Mendecki (1997))
VA = M/(c3A ) = M2 /(c3 GE)
( 2-1)
where c3 =2 (a scaling factor) ,
M is the seismic moment,
A is the apparent stress,
G is the shear modulus and
E is the released energy.
Mendecki (1997) quantifies accelerated deformation through accumulated Apparent Volume,
VA. , however, this project recognises accelerated deformation through the rate of change in
Apparent Volume. Accumulated Apparent Volume, VA., may create the impression of
accelerated deformation through an increased event rate with all events being similar as for
Apparent Volume, whereas this project wanted to recognised Apparent Volume rate of change
as unique input evidence. A 'short term / long term' Apparent Volume ratio is calculated. The
ratio is calculated at each event occurrence and is the median of the Apparent Volume for the
last 10 events versus the median for the last 50 events.
As before with event rate, Apparent Volume as input evidence towards the hypothesis of a RE
incident during the next shift has to be quantified.
Table 2-2 Apparent Volume - quantified input evidence parameters
Parameter
p(E|H)
p(E|~H)
Ryes
Rno
Apparent Volume
0,5
0,14
2,0
0,5
Figure 2.2 shows the same events as in Figure 2.1, but now showing the rate of change in the
Apparent Volume as observed in a short-term long-term ratio. The red contour implies a 'short
term / long term' ratio of two.
12
Figure 2.2 Rate of change in the Apparent Volume as observed in a short-term long-term
ratio
2.1.5 Energy Index as an risk indicator of possible rock burst damage
The parameter Energy Index was also initiated by Mendecki, (1997) as a way of comparing
radiated seismic energy from events of similar seismic moments. An event with Energy
Index > 1 would suggest a higher than average shear stress at its location. The opposite
applies when Energy Index is < 1.
As before, for event rate and Apparent Volume, the Energy Index parameters are quantified as
follows for the Bayes’ Theorem:
Table 2-3 Energy Index - quantified input evidence parameters
Parameter
p(E|H)
p(E|~H)
Ryes
Rno
Energy Index
0,53
0,15
0,15
-1,0
As in Figure 2.1 and Figure 2.2, Figure 2.3 shows the rate of change in Energy Index as
overlain on the mining operation. The scale ranges from -3 for blue, to +0,5 for light green. It
can be inferred from this figure that there is some increase in the driving stress of the previous
10 events versus the longer term average, in the top part of the panel.
13
Figure 2.3 As in Figure 2.1 and Figure 2.2 but this time the rate of change in Energy
Index is overlain on the mining operation
2.1.6 Peak ground motion as an risk indicator of possible rock burst
damage
In GAP 714 (Brink et al.,2002), a model was presented of peak ground velocities in the near to
far field and applied it to data from a deep level mine. It was assumed that future seismicity is
likely to be similar to historical seismicity. A single equation was developed to express peak
velocity.
The following were assumed:

All seismic events occur on Brune-type circular slip zones in plan around each
event location.

The rock mass is elastic and homogeneous. Site effects and amplification at the
skin of the stope are neglected.
Models of seismic sources generally consider strong ground motion either in the near field or in
the far field. Employing these models the peak velocity can be expressed as a single equation:
v = (VS/G) * (r0/R/)
( 2-2)
for R>=r0
where VS = shear-wave velocity,

 = static stress drop,
R = hypocentral distance,
r0 = source radius
and
G = modulus of rigidity
(Described by McGarr (1991) and also applied by Brink et al.,(2002))
14
2.1.6.1 Implementation of peak ground motion as risk parameter
Figure 2.4 shows an example of the implementation of peak ground motion as a risk parameter.
A grid with a resolution of 25 m was placed over the selected area. At each grid element all the
seismicity in the area is evaluated for its respective peak ground motion at that point. An
arbitrary threshold (say, 1 m/s) is set. The number of times that a grid element experiences a
ground motion of more than the threshold is logged. It is clear that the highest values are
experienced at the current face position. It is assumed that each grid element is, in fact, in an
excavation, and a multiplication factor of 3 is used to allow for site amplification.
The peak ground motion and the number of times that an excavation experience significant
ground motions, can be linked to physical damage experienced in the excavation (Andersen et
al, (1999)).
Peak particle velocity (PPV) count, does not necessarily relate to the potential for experiencing
a high PPV at a specific grid position, other than that the return time between incidents of high
ground motion is shorter. The PPV count is used in this evaluation as an indicator of the
vulnerability of the excavation to ground motion. A larger number of large ground motions
should relate to the extent of seismic related damage in an area. A higher degree of earlier
damage makes an area more vulnerable to the next higher level of ground motion. In Figure 2.4
the centre of the mini-longwall experienced the largest number of ground motions larger than
0,2 m/s. The power law distribution of PPV counts justifies the expectation that this area also
would have experienced more large ground motions.
Figure 2.4 A plot of the number of times during the previous 30 days that an estimated
peak particle velocity of 0,2 m/s was exceeded
For peak particle velocity the project is not attempting to recognise a change in the number of
PPV counts larger than a preset threshold, and rather used the absolute number of PPV counts
exceeding the threshold during the previous 30 days.
Table 2-4 Input values for peak particle velocity counts larger than 0,2 m/s for the
application of Bayes’ Theorem
Parameters
p(E|H)
p(E|~H)
Ryes
Rno
PPV count
0,8
0,5
3
0
15
3
Evaluation of the risk assessment software.
The objective of this project was to provide clear documentation of actual application of a rock
engineering risk assessment tool. This could not be achieved in real time because of severe
limitations in access to real-time seismic data on mines, as well as system interfacing to an
operational mine’s database for rock engineering parameters. The project conducted a
demonstration on an ‘off-’line’ case study with two years seismological data as input. A subset
of the results of this continuous risk assessment process is captured in Appendix A. The
analysis and display software, as well as the seismic data are included in the attached CD
(Appendix B).
Although a grid resolution of 10 m x 10 m was used, event source size and location accuracy
did not allow the process to uniquely link an event to a single grid position. A zone of influence
around the event location was used. This zone of influence could be determined, for example,
using source dimension, apparent volume or just a 1/(distance to location) relation. For the
purpose of this proof of concept a fixed zone of influence of 100 m diameter was used.
All the earlier mentioned evidence, namely event rate, Apparent Volume, Energy Index and
ground motion is combined in estimating the probability of experiencing a rock related incident
at a specific grid element during the next shift.
3.1
Incorporating rock mechanics parameters
The earlier-mentioned input parameters were solely based on the 'dynamic' information and did
not take into account the vulnerability of the excavation as described by parameters such as
ERR, support quality and local ground condition. To incorporate the 'static' risk information, it
was necessary to define the total working area, i.e. the active panels and the back areas. The
total working area was defined in Figure 3.1. In this example the back areas were only
considered in terms of the exposure of people. No specific risk attributes were allocated to any
grid position.
Figure 3.1 Total working area to be considered in terms of risk to the underground
personnel. The area includes the active panels and the back areas
The next step was to specify the input parameters to specific active panels. Panels are typically
known by name rather that geographic position. A panel name and its associated attributes had
to be linked to a specific geographical area. The output of this process is shown in Figure 3.2.
Every panel, in terms of its geographical position, was now linked to the RE database. The
16
attributes described the vulnerability of the excavation to large ground motion and also include
the exposure at a particular time of day as well as the more general 'Quality of information' (as
described by Brink et al., 2002).
Figure 3.2 The process of linking a geographical area of an active panel to a set of
attributes
3.2
A case study on continuous risk assessment
The probability approach leads to a large dynamic range in output levels and the linear display
may give the impression of a insignificant risk in some areas. The risk levels as displayed in
Appendix A uses a logarithmic scale.
Appendix A depicts a series of events from 16 February to 2 March. Each figure represents a
risk assessment based on the input evidence immediately prior to the event shown. The event
parameters as well as the located position are shown. The scale on the right is used throughout the presentation. It is a logarithmic scale where full-scale relates to a 50% probability, with
average weighting factors, that a rock-related incident will take place within the subsequent
eight hours. The arrow indicates the position of the last event. The event parameters are shown
on the figure.
It can be noticed that distant events have a negligible effect on the probability of the RE incident
in a particular area. In this application, an event more than 100 m away can only have an
influence in terms of peak ground motion which in turn will only be significant should a major
event occurs.
By following the event sequence in Appendix A, it will be noticed that the first significant event
occurred 22 February at 02:35. This event was not preceded by any increase in the assessed
risk. A contributing factor to the low prior risk levels is the fact that this area is classified as back
area with a three times lower exposure of people than in the stope areas.
Within a day the risk in the top panel started to increase, to be followed by a Magnitude 2 event
about fifteen hours later during the morning shift (24 February at 10:33). The risk in the top
panel stayed high and extended to the neighbouring panel, where another event (Mag. = 1.7)
occurred during the blasting time (24 February at 16:35). Another event (Mag. = 1) occurred
during the next blasting time (25 February at 16:20). After a further two days the risk on these
panels had return to normal. This sequence is just a ‘snap shot’ from of the two year’s data and
is very representative of many similar sequences.
17
The input parameters to the expert system were primarily the temporal behaviour of the seismic
event parameters. Accepting this, it therefore follows that should all event parameters be
maintained but the event occurrence times be randomised, that the information contained in the
temporal behaviour should be scrambled. The next sequence in Appendix A contains the same
events, therefore the same event characteristics but now with a random time added and/or
subtracted (+/- 120 hours) per event. The objective is to demonstrate the validity of the risk
assessment process applied in the previous sequence.
By scanning through this randomised sequence and comparing it with the previous sequence,
the reader will recognise that the precursive risk indication has totally disappeared. This is a
substantive proof of the validity of the engaged process.
Over a time window of one year more than 80% of the larger events (Mag > 1) in the immediate
vicinity of operating panels in the case study area, were preceded by a significant increase in
the risk levels as determined by the expert system process. Similarly, less than 40% of all high
risk levels did not manifest into a relatively large event. These levels were obtained subjectively
by watching the risk display over this period of time. An objective and quantified measure of
determining success rate was not proposed in this project. Absolute quantification of the
success rate should be possible, but then again, the implication is that this concept is being
presented as a seismic prediction tool, rather than a methodology to assess RE risk.
4
Conclusions
The concept of integrated RE risk assessment was described and an initial version of the
software was developed and used in a case study. This report can not fully describe or
demonstrate the application of this code - in particular the dynamic display of information. For
this purpose a series of output screen display were captured and presented in Appendix A.
The case study included a continuous assessment of RE risk on a mining section on the VCR in
Far West Rand. A significant majority of the larger events (Mag > 1) in the immediate vicinity of
operating panels were preceded by a significant increase in the assessed risk levels.
The deliverable of this project was an evaluation of a prototype software package. The project
did not succeed in achieving a real-time evaluation because of a number of system interfacing
constraints, specifically to the seismic input parameters.
It has to be stressed that the objective of the project was not to provide a tool for the prediction
of large seismic events, but to provide an integrated approach towards the dynamic assessment
of the overall RE risk at a particular place in the mine; at a particular time; and with cognisance
of the exposure of the underground worker to the inferred RE risk.
Evaluation of the methodology proved to be difficult. It can only be properly evaluated through
extended use in an operational environment. The original proposal intended to have the code
tested in real-time on a Far West Rand mine and on a relatively deep platinum mine. It was not
possible, and would have been rather ambiguous to define specific case studies and to claim a
percentage success rate as in the case of prediction.
The project can be criticised in that the non-seismic aspects of RE risk assessment did not
contribute to the case study evaluation. The interfacing problems to mine RE data were the
cause of this valid criticism. However, the concept of using changing input evidence that could
influence the probability of an output hypothesis being true was well proven.
A future software implementation of the proven concept should be achievable by a professional
software developer, without this person needing to understand the basic principles employed in
this demonstration/evaluation version. The important outstanding aspects of this potentially
useful tool are mainly the real-time data and user interfacing. It is recommended that such an
investment should be made.
18
5
References
Andersen, L. and Daehnke. A. 1999. Final project report for internal and unpublished project,
Deepmine 4.1.2.
Brink, A.v.Z., Hagan, T.O., Spottiswoode, S.M., Malan, D.F. 2000. Survey and assessment of
the techniques used to quantify the potential for rock mass instability, SIMRAC Report GAP
608, Department of Mineral and Energy Affairs, South Africa.
Brink, A.v.Z., Hagan, T.O. 2002. Software tool for managing rock related hazards in South
African mines, SIMRAC Report GAP 714, Department of Mineral and Energy Affairs, South
Africa.
Brink, A.v.Z. and Newland A.R. 2001. Automatic Real Time Assessment of Windblast Risk,
ACARP project C8026, Australian Coal Association Research Program.
Lenhardt, W.A. 1998. Zur Risikobewertung bergbauinduzierter Seismizität. Felsbau, 16, Nr. 1,
48-55.
McGarr, A. 1991. Observations constraining near-source ground motion estimated from locally
recorded seismograms, J. Geophys. Res., pp 16495-16508.
Mendecki, A.J. 1997. Quantitative seismology and rockmass stability, in Seismic Monitoring in
Mines, A J Mendecki (Ed.), Chapman and Hall, Cambridge, 1997.
Menoni, S. Petrini, V. Zonno, G. 1997. Seismic risk evaluation through integrated use of
geographical, ENV4 - CT96 – 0279, Istituto di Ricerca sul Rischio Sismico, Milano, Italy.
Milev, A.M., Spottiswoode, S.M. and Noble, KR. 1995. Mine-induced Seismicity at East
Rand Proprietary Mines, Technical Note, Int. J. Rock Mech. Min. Sci. & Geomech. Abstr., v32,
no6, pp 629-632.
Naylor, C. 1998. XMaster® : An Easy-to–use expert system shell for Windows. Expert update:
The Bulletin Of The British Computer Society's Specialist Group On Knowledge-Based Systems
And Applied Artificial Intelligence, Autumn 1998 Volume 1 Number 2.
Naylor, C. 2000. Developer's guide, XMaster® , published by Chris Naylor Research Limited.
19
Appendix A
Technology transfer of the risk
assessment tool for rock related risk
issues
Project Number:
SIM 03 02 01
A.v.Z Brink
Date
time
x cor
y cor
Mag
16 Feb
03:31:34
29403
-39189
-0.2
Comment 1. The arrow indicates the position of
the last event. The event parameters are shown
on the figure. The risk display is the
assessment immediately prior to the event.
The scale on the right is used through-out this
presentation. It is a logarithmic scale where fullscale relates to a 50% probability, with average
weighting factors, that a rock-related incident
will take place within the next 8 hours.
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
16 Feb
16:50:14
29358
-39203
0.8
16 Feb
18:03:48
29388
-39172
0.0
20
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
17 Feb
10:56:05
29359
-39193
0.3
18 Feb
16:17:57
29352
-39158
0.1
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
20 Feb
07:36:21
29337
-39193
0.4
20 Feb
16:45:17
29279
-39222
0.4
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
20 Feb
16:53:52
29435
-39169
0.3
21 Feb
16:33:27
29181
-39275
0.0
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
21 Feb
18:29:20
29284
-39215
0.2
22 Feb
01:26:44
29376
-39198
-0.2
21
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
22 Feb
01:28:38
29505
-39210
-0.7
22 Feb
01:37:40
29570
-39165
-0.4
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
22 Feb
01:43:56
29549
-39187
-0.2
22 Feb
01:45:39
29582
-39354
-0.2
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
22 Feb
02:16:59
29146
-39279
-0.4
22 Feb
02:35:09
29262
-39250
1.1
Comment 2. A magnitude 1.1 event happened
outside the area designated as working area,
therefore with low exposure of people and
subsequent low risk. No prior indication of a rockrelated incident is noticed.
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
22 Feb
09:59:18
29271
-39262
0.8
22 Feb
11:36:04
29284
-39261
0.0
22
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
23 Feb
15:52:18
29315
-39195
0.2
23 Feb
23:44:05
29298
-39269
-0.1
Comment 3. The top panel is beginning to experience
a high risk
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
24 Feb
02:52:49
29353
-39203
0.2
24 Feb
10:33:07
29292
-39202
2.0
Comment 4. A Magnitude 2.0 event occurred on the
panel where an increased risk could be observed
about 15 hours earlier.
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
24 Feb
16:25:57
29321
-39167
0.1
24 Feb
16:27:58
29273
-39290
0.3
23
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
24 Feb
16:35:03
29379
-39166
1.7
25 Feb
04:10:36
29237
-39278
0.5
Comment 5. The second panel suffers a Magnitude
1.7 event after a shift of the higher risk indication onto
this panel.
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
25 Feb
07:42:00
29205
-39263
-0.5
25 Feb
10:03:17
29461
-39173
0.7
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
25 Feb
16:20:47
29346
-39184
1.0
25 Feb
16:22:51
29320
-39201
0.0
Comment 6. Another larger event (Mag 1) affects the
second panel a day later. Again it happened during
the blast.
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
25 Feb
16:34:59
29459
-39144
0.1
25 Feb
18:28:26
29329
-39196
0.0
24
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
26 Feb
16:39:48
29309
-39200
0.1
26 Feb
16:41:47
29281
-39209
0.0
Comment 7. After another two days the risk on these
panels had return to normal.
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
26 Feb
21:57:25
29292
-39199
0.8
27 Feb
16:25:38
29169
-39308
-0.1
Date
time
x cor
y cor
Mag
02 Mar
22:59:00
29380
-38743
0.6
25
Random series
• Similar events, but now with a random time
added/subtracted (+/- 120 hours) per event.
• The objective is to demonstrate the validity of
the previous sequence.
Date
time
x cor
y cor
Mag
15 Feb
20:21:59
29317
-39212
0.4
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
16 Feb
05:08:59
29440
-39177
0.3
17 Feb
10:11:05
29435
-39169
0.3
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
17 Feb
13:35:54
29358
-39203
0.8
17 Feb
10:11:05
29435
-39169
0.3
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
17 Feb
18:17:07
29284
-39261
0.0
18 Feb
03:47:46
29582
-39354
-0.2
26
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
18 Feb
08:48:38
29359
-39193
0.3
18 Feb
13:43:01
29277
-39329
0.3
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
19 Feb
00:10:42
29281
-39342
0.3
19 Feb
09:36:18
29271
-39262
0.8
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
19 Feb
16:56:59
29181
-39275
0.0
19 Feb
18:24:34
29505
-39210
-0.7
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
20 Feb
03:30:48
29315
-39195
0.2
20 Feb
06:55:08
29284
-39215
0.2
27
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
20 Feb
10:48:35
29279
-39222
0.4
20 Feb
18:09:10
29292
-39202
2.0
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
21 Feb
04:06:44
29329
-39196
0.0
21 Feb
18:07:35
29205
-39263
-0.5
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
22 Feb
01:50:33
29292
-39199
0.8
22 Feb
11:31:39
29321
-39167
0.1
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
23 Feb
02:30:00
29376
-39198
-0.2
23 Feb
06:25:25
29237
-39278
0.5
28
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
24 Feb
03:10:20
29379
-39166
1.7
24 Feb
07:09:50
29273
-39290
0.3
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
24 Feb
07:32:07
29262
-39250
1.1
24 Feb
10:01:57
29146
-39279
-0.4
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
24 Feb
10:48:58
29320
-39201
0.0
24 Feb
20:05:04
29346
-39184
1.0
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
24 Feb
22:34:48
29298
-39269
-0.1
25 Feb
04:11:31
29549
-39187
-0.2
29
Date
time
x cor
y cor
Mag
Date
time
x cor
y cor
Mag
25 Feb
05:08:04
29309
-39200
0.1
26 Feb
01:52:38
29353
-39203
0.2
30
Appendix B
The project output is compiled on a CD, comprising all the evaluation coding, input data and
results. The CD includes the following coding and data files:
VBA code to be used with Excel.doc
A listing of the VBA code used
Presentation risk.ppt
A PowerPoint
Appendix A
Sim 03 02 01 data.xls
A data set of seismic activity in a section
over two years
31
presentation
similar
to