Drill monitoring results reveal geological conditions in blasthole

International Journal of Rock Mechanics & Mining Sciences 78 (2015) 144–154
Contents lists available at ScienceDirect
International Journal of
Rock Mechanics & Mining Sciences
journal homepage: www.elsevier.com/locate/ijrmms
Technical Note
Drill monitoring results reveal geological conditions in blasthole
drilling
Peter Hatherly a,n, Raymond Leung a, Steven Scheding a, Danielle Robinson b
a
b
Australian Centre for Field Robotics, The University of Sydney, NSW 2006, Australia
Rio Tinto Technology & Innovation, Australia
art ic l e i nf o
Article history:
Received 22 October 2014
Received in revised form
14 May 2015
Accepted 19 May 2015
Keywords:
Monitor-while-drilling (MWD)
Rock mass characterisation
Geophysical logging
Open pit mining
Blasting
1. Introduction
In open pit mining, monitor-while-drilling (MWD) systems
which monitor performance factors such as rate of penetration
(RoP), torque (τ), weight on bit (WoB) and rotary speed (ω) are
becoming standard features on the blasthole drill rigs supplied by
equipment manufacturers. MWD measurements enable drill
automation1 and can be used to monitor the health of these major
items of equipment. While it has also long been recognised that
through MWD data, geological conditions can be revealed,2–4
MWD data is rarely used for geological characterisation in activities such as blast design.
Part of the reason for this lies with the difficulty in keeping
MWD systems operational in the hostile mining environment.
Unless they are integral to specific mining tasks (e.g. automation),
MWD systems are frequently turned off or not fully operational.
However, were there a better appreciation of how well MWD data
reflect changes in the geology, mining companies might begin to
make use of MWD data for geological and geotechnical purposes.
In making comparisons between MWD and geological data,
difficulties arise in obtaining geological information for comparison with the MWD data. Blastholes have relatively large diameters (4200 mm) and are drilled using rotary and percussion
methods. Drill chips do not provide adequate information and if
geological and geotechnical models developed from broadly
n
Corresponding author.
E-mail address: [email protected] (P. Hatherly).
http://dx.doi.org/10.1016/j.ijrmms.2015.05.006
1365-1609/& 2015 Elsevier Ltd. All rights reserved.
spaced (100 m plus) exploration drilling are used, these are not
sufficiently precise for comparison with blastholes which are typically drilled at centres of about 10 m. Even when cores from
nearby cored holes are available, these are not adequate because
the cores are not from the actual blastholes and in any case, once
cores are removed from the ground, they are no-longer under insitu conditions and it is not possible to capture the influence of insitu stresses and fractures. Given the difficulties with all of these
approaches, geophysical logging conducted within the blastholes
probably provides the best opportunity for obtaining precise
geological information from within the blastholes drilled with rigs
equipped with MWD capabilities.
In this technical note we provide results for a study undertaken
at an open cut coal mine in the Hunter Valley of NSW, Australia.
The MWD data were obtained using Caterpillar's Aquilas drill
monitoring system. RoP measurements are assessed as well as the
specific energy of drilling (SED).5 Geophysical logging was then
conducted in the same blastholes and the results analysed to establish the down-hole lithology and rock quality.
At the mine where this work was undertaken, the main incentive for the study was to investigate whether MWD data could
identify the top of coal seams and allow drilling to be halted, thus
preventing unintended blasting of coal and associated problems
with coal fragmentation and dilution. A description of these results and a new MWD measure called Modulated Specific Energy
(SEM) is introduced in Ref. 6. This approach is well suited for
picking the changes in MWD performance when coal seams are
encountered.
P. Hatherly et al. / International Journal of Rock Mechanics & Mining Sciences 78 (2015) 144–154
The more general blast design problem and the potential for
MWD to reveal changes in overburden rock quality are now considered. Of particular interest is the association between MWD
parameters and sonic velocity. Velocity has been used as a proxy
for rock strength by many authors.7–12 A direct correlation between MWD data and sonic velocity can therefore be expected.
Also of interest was any association between the MWD data and
the rock lithology revealed by natural gamma logs (sandstones
versus siltstones and claystones) and density logs (clastic rocks
versus coal and carbonaceous layers containing mixtures of coal
and clastic sediments).
Previous blasthole studies employing geophysical logs are described in Ref. 13 where geophysical logs were used to indicate the
locations of coal seams and Ref. 14 where density logs were used
to assist with the application of pattern recognition techniques for
rock characterisation using MWD data in an open pit coal mine.
145
Neither of these studies extensively explored the correlation between MWD parameters and geological conditions revealed by the
geophysical logs.
In the case of the MWD techniques and geophysical logging
employed by the petroleum industry, that work is not directly
relevant to the rock characterisation problems in open pit mines.
In petroleum, holes are hugely expensive, drilled to depths of
many thousands of metres and are widely spaced. In these very
deep holes, the use of drilling fluids, the interaction between the
bit and the rock, bit wear and the abrasivity of the strata are major
considerations. In open pit mining, blastholes are cheap, drilled
with air and drilled to depths of no more than a few tens of metres
in a matter of minutes. As already noted, they are also closely
spaced. In mining, bit–rock interactions and variations in bit wear
down-hole and from hole-to-hole are not major concerns when
characterising rocks.
2. Site description
Increasing sandstone
N
Fig. 1. Blasthole pattern for the bench at the trial site. Holes are in seven rows (B to
H). The holes shown by black in-filled circles were drilled through to the base of the
coal seam. These were geophysically logged with natural gamma, caliper and
density tools. In addition, sonic logs were run in the holes which are labelled and
shown by red in-filled circles. For the sonic logs to be run, it was necessary to fill
these holes with water. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
The site for the trial is an open pit coal mine operating within
the Jerrys Plains Subgroup of the Whittingham Coal Measures and
part of the Permian sequence of the Sydney Basin.15 It was formed
in an upper to lower delta plain terrestrial environment and the
interburden rocks between the coal seams consist mainly of lithic
sandstone, siltstone and conglomerate. The thirteen named coal
seams frequently split and provide numerous mining opportunities. Carbonaceous siltstones and tuffs also occur throughout the
Subgroup.
Fig. 1 shows a map of the bench at the trial site. Here, an interburden approximately 24 m thick overlies a 2–3 m thick coal
seam, depending on the thickness of a mid-seam split. The interburden is mainly sandstone and siltstone and as shown by the
photographs in Fig. 2, siltstones in the north grade into sandstones
in the south. At this location, a layer of carbonaceous siltstone
which can also contain a minor split of the coal overlies the main
coal seam.
One-hundred forty-four blastholes were drilled by experienced
operators using CAT Reedrill SKS rotary drill rigs in preparation for
a normal production blast. There were seven north–south rows of
blastholes 7.5 m apart, with the blastholes within each row at
10 m centres. Most holes were drilled to the top of the seam with
the driller ceasing drilling at the expected coal seam depth and
also according to any observed changes in the RoP and the colour
of drill chips. As an exception to this normal practice, blastholes
were deliberately drilled through to the base of the seam in five
east–west rows. This provided the opportunity to properly establish the seam depth and to obtain geophysical and MWD data
through the coal.
The diameter of the blastholes was approximately 26 cm. For
the 35 blastholes drilled through the coal, geophysical logs were
obtained by a geophysical contractor using conventional slimline
tools normally used in exploration boreholes. Geophysical measurements were taken with a combined density, natural gamma
and caliper tool (7 cm diameter tool pushed against the wall of the
hole). Full waveform sonic (FWS) logs (5 cm diameter tool centralized within the blasthole) were also obtained in three holes in
each of the five rows, see Fig. 1. To obtain the sonic logs, these
holes needed to be first filled with water because sonic logs require fluid coupling to transfer signals between the tool and the
blasthole walls.
As discussed below, there are differences in the hole depths
reported in the MWD and geophysical data. This is due to the
practice of the drill operator to set the zero depth at the point
when the drill was set into drilling mode point, approximately
0.3 m above the surface of the bench. In the case of the geophysical
146
P. Hatherly et al. / International Journal of Rock Mechanics & Mining Sciences 78 (2015) 144–154
Siltstone
Sandstone
Fig. 2. Photographs of the highwall immediately to the west of the test site
showing the strata present within the bench. Siltstones with thin sandstone beds
are mainly present to the north (top photograph) and thick lithic sandstones are
present to the south (bottom photograph). Remnants of pre-split blast holes are
present in the sandstones. These are not present in the weaker siltstones.
logs, the geophysical logging operator visually set the zero depth
point when it was judged that the top of the tool was level with
the surface of the bench. While slight differences between depth
measurements for the two geophysical tools in each hole are
likely, much greater discrepancies exist between the geophysical
and MWD measurements.
3. Geophysical logging data
The raw density, natural gamma and caliper logs were obtained
at 1 cm intervals up the blastholes. These data were first filtered to
remove high frequency noise using a 29-tap Hanning filter and
then resampled at 10 cm increments in order to match the sampling interval of the MWD data. Fig. 3a and b show examples of
these logging data obtained in blasthole B15, after filtering and
resampling.
For the density measurements, short spaced (15 cm source to
detector separation) and long spaced (31 cm separation) measurements were available. The long spaced measurements were
used in this study because they are less susceptible to variations in
the immediate blasthole wall. Calibration to units of t/m3 was
provided by the logging contractor. The logging contractor also
provided calibrated natural gamma measurements in standard
units of API (American Petroleum Institute). For the caliper tool,
the measurements were taken using a single arm mechanical
caliper.
In the case of the sonic logging, full waveform sonic data were
acquired at 2 cm intervals and measurements were taken by two
receivers separated by 30 cm. Onsets (transit times) of the first
arrivals (P-waves) recorded at each receiver were picked using an
interactive seismic event picking programme. As can be seen in
the waveforms recorded in blasthole B15 in Fig. 3(c) and (d),
identification of the P-wave onsets was not always possible, particularly when the sonic velocities in the strata in the were low.
This problem arose because the large diameter of the blastholes
compared to the sonic tool created a situation whereby pressure
(tube/Stoneley) waves propagating within the water column in the
blasthole could arrive at the receivers before the arrival of the
sonic (P-waves) refracted up the blasthole wall. This was a problem mainly in the coal. There are also intervals within the holes
where the amplitudes of the P-wave arrivals are very low and it
was not possible to identify their onset. In both of these situations,
P-wave velocities could not be determined and it is for these
reasons, the velocity measurements shown in red in Fig. 3b are not
available for the full length of the blasthole. As with the density,
gamma and caliper logs, the sonic data were resampled to 10 cm
intervals.
Examination of the geophysical data shows the key features in
the geology and some issues with the data. In the density data
(Fig. 3a), the locations of the coal seams are indicated as intervals
with low density. From this and the other density logs, coal was
identified whenever the density was less than 1.85 t/m3 and
carbonaceous material was taken to be present whenever the
densities were between 1.85 and 2.2 t/m3. All other material with
densities greater than this was inferred to be clastic (sandstone
and siltstone). In blasthole B15, the main coal seam occurs between 21 and 24.8 m, with a mid-seam band of carbonaceous
material occurring between 23 and 23.7 m. A minor coal seam is
also present at the very bottom of the hole.
Outside of the intervals of coal and carbonaceous materials
where clastic rocks are present, comparison of the density and
caliper logs (Fig. 3a) indicates some degree of correlation between the logs. Such a correlation should not occur and is possibly due to the influence of the large drillhole diameter. The
density logs were therefore only used to indicate the locations of
the coal seams and carbonaceous layers, and not for any other
purpose.
The natural gamma log (Fig. 3b) shows that the coal seams are
typified as regions with low values of about 50 API, while in the
clastic section of the blasthole, the natural gamma measurements
are higher. When they are about 100 API, lithic (dirty) sandstones
can be inferred to be present and when the values are about 150
API or higher, siltstones are present – i.e. in the siltstones there are
more clay minerals and less quartz. On the log for B15, it is notable
that from a depth of about 14 m down to the top of the coal seam,
there is a gradual increase in the natural gamma measurements.
Gradual increases such as this indicate that the transition from
sandstone to siltstone is gradual and that there is no sharp
boundary between these two rock types.
In the clastic section of the blasthole where it has been possible to obtain P-wave sonic data, it can be seen that the P-wave
velocities range between about 2 and 4 km/s. The lower velocities
are found in the materials shown by the natural gamma log to be
siltstones. The higher velocities occur in the sandstones. Below
14 m in this hole, within the transitional zone from sandstone to
siltstone, there is a matching decline in velocity. Other features
are also present in the sonic data. For example, between 8.2 and
8.6 m, in the sandstone section of the blasthole, the velocity is
low (about 2.2 km/s). Such lower velocities are presumably due
to the presence of fractures within the sandstone. At 13.4 m,
there is a situation where the sonic velocity is higher while the
natural gamma is lower and the density is higher. At this point, a
P. Hatherly et al. / International Journal of Rock Mechanics & Mining Sciences 78 (2015) 144–154
147
Fig. 3. (a)–(d) Geophysical logs for blasthole B15-the top of the coal seam is at 21 m. (e) to (h) MWD parameters-features at 12 m and 24 m depth are due to rod changes and
there are other disruptions to drilling at 20.5 m and 23 m. (i) SED derived from the drilling parameters with effects of rod changes and drilling disruptions removed. (For
interpretation of the references to colour in this figure, the reader is referred to the web version of this article.)
148
P. Hatherly et al. / International Journal of Rock Mechanics & Mining Sciences 78 (2015) 144–154
Table 1
Nominal drill operating conditions.
Rotary speed
(rpm)
Process mean μ̄
143
9.6
Process s.d. σ̄
Max observed s.d. 32.5
Torque
(kN m)
RoP (m/s)
WoB (kN)
6.7
0.62
1.93
0.045
0.016
0.089
1040
70.1
201
hard siderite band is likely to be present in the blasthole wall.
Siderite bands are common at many Australian coal mines.
4. MWD data
RoP (m/s)
RoP (m/s)
The rotary drilling rigs deployed at the site provide real-time
measurement of several mechanical signals that are commonly
used to monitor drilling performance. Of interest to this study are
the four variables rotary speed, WoB, torque and RoP. These are
measured in units of revolution/minute, kN, kN m and m/s respectively. Although they naturally exist as continuous-time signals in the analogue world, they were presented by the logging
system as discretised sequences (quantised in depth, not time) at
0.1 m resolution.
For rock mass characterisation using MWD data, the general
rule governing the operation of the rotary drill is that the control
parameters (essentially, the rotary speed and WoB) should remain
constant and stable as far as possible. When this is achieved, it has
been shown that variation in torque and RoP may reveal differences in geological conditions.2 The expectation is that RoP will
increase as the drill penetrates softer material. While these observations may hold true over wide-ranging conditions, any inference rules developed between MWD measures and rock quality,
if such connections can be established, will likely hold in a relative
sense, and remain strictly valid only under a specific set of operating conditions. The typical operating range of the rotary drills
used in this study is shown in Table 1.
Fig. 3e–h show examples of the raw MWD sequences encountered in the same blasthole as the geophysical data in
Fig. 3a–d. These illustrate that noise is ever present in real signals. With reference to the WoB and rotary speed, the major
departures from the ‘constant control input’ assumption occur at
periodic intervals roughly every 12 m and are due to rod changes.
When this occurs, transient responses can arise due to drilling
interruption and changes in drilling dynamics. In practice, the rod
change artefact can be detected by a diminished RoP and a drop
in drill-pipe air pressure, and was corrected by replicating
(holding out) the previous sample not affected by the distortion.
As discussed in Appendix A, other outliers in the MWD data were
also removed.
While RoP provides a primary indication of drill performance, it
is dependent on the other measures. To remove this inter-dependence, the SED was also calculated. Fig. 3i shows the SED values for blasthole B15.
Velocity (km/s)
SED (GJ/m3 x 0.1)
SED (GJ/m3 x 0.1)
Natural gamma (API)
Natural Gamma (API)
Velocity (km/s)
Fig. 4. (Top) Scatter plots showing variations in RoP with natural gamma and sonic velocity. (Bottom) Scatter plots showing variations of SED with natural gamma and sonic
velocity.
P. Hatherly et al. / International Journal of Rock Mechanics & Mining Sciences 78 (2015) 144–154
5. Analysis—MWD versus geophysics
Once the data from the 35 blastholes containing the geophysical logs were matched in depth with the MWD results (see Appendix B), the relationship between them and the drilling data
could be explored. In all, there was 1056 m of drilling with MWD
and geophysical data. After checking on the natural gamma and
MWD data, it was found that 4797 data points were acceptable.
For the 15 holes with sonic logs, it was found that there were 1535
points available, of which 47 were in coal (density less than
1.85 t/m3) and 60 in carbonaceous material (density between 1.85
and 2.2 t/m3). The remaining points were within clastic rocks
(sandstones and siltstones).
5.1. X–Y plots
Fig. 4 shows X–Y scatter plots of RoP and SED against the natural gamma measurements and sonic velocity. With regards to the
scatter plots involving sonic velocity, it is clear that there is a trend
whereby RoP decreases with sonic velocity and SED increases. This
is as expected. On both plots, there is also a small cluster of points
with low velocity which lies outside the main grouping of points.
These are from coal and are further discussed below.
In the plots showing the MWD relationships against natural
gamma responses, two groupings are again evident with the cluster
of points with the lower natural gamma response again representing
coal and the majority of the points representing the clastic rocks.
149
Given the number of points that are contained in these scatter
plots, it is difficult to properly identify the trends involving the
majority of points. These trends can be more thoroughly explored
by separating out the responses for each of the main rock types
and undertaking a statistical analysis of the individual
relationships.
As mentioned above, coal and carbonaceous materials were
identified on the basis of the density response. To separate the
sandstones from the siltstones, an assessment was made of the
natural gamma logs for each hole and a sand–clay baseline was
established at a value of about 125 API, depending on the characteristics of the individual natural gamma logs. Points with values
below the baseline were assigned to the sandstone category and
points with values above were labelled siltstone. This leads to a
hard cut-off between siltstone and sandstone which is inconsistent
with the gradual change from sandstone to siltstone evident in
Figs. 2 and 3. However, it is useful to be able to characterise the
rocks as being one of these two classes while acknowledging at the
same time the continuum that exists between them.
The approach taken to generate X–Y plots in the form of heat
maps between the MWD and geophysical parameters for each rock
type is explained in Appendix C. It allows the responses from the
disproportionately low numbers of samples from coal and other
carbonaceous materials (especially in the case of the sonic data) to
be compared with those from sandstone and siltstone. This requires the conditional distribution of points with respect to the
X–Y variables to be computed for each class (rock type).
Coal
Coal
Silt
RoP (m/s)
RoP (m/s)
Carb
Carb
Sand
Silt
Sand
Natural gamma (API)
SED (GJ/m3 x 0.1)
Sand
Silt
Carb
Coal
SED (GJ/m3 x 0.1)
Velocity (km/s)
Sand
Silt
Carb
Coal
Natural Gamma (API)
Velocity (km/s)
Fig. 5. Heat maps showing variations in RoP and SED with natural gamma and velocity measurements. Distributions for each of the major rock types were independently
derived and then combined to provide these representations. The rock types associated with each of the features in the maps are labelled.
150
P. Hatherly et al. / International Journal of Rock Mechanics & Mining Sciences 78 (2015) 144–154
Table 2
Correlation coefficient between MWD measures (X) and geophysical attributes (Y) as a function of outlier rejection ratio.
Outlier rejection ratio, r
Y
X
0%
5%
10%
15%
20%
25%
50%
Velocity
Velocity
RoP
SED
0.643
0.666
0.725
0.726
0.761
0.760
0.780
0.805
0.786
0.836
0.789
0.859
0.802
0.924
The heat maps corresponding to the scatter plots in Fig. 4 are
shown in Fig. 5. It is now possible to see with greater clarity how
the MWD and geophysical parameters are related.
In the case of the very good relationships with the sonic velocity, the sandstones are associated with slower drilling and increased drilling energy. The continuum from sandstone to siltstone
and the easier drilling conditions is clearly revealed. Coal provides
the easiest drilling and is represented by clusters that lie outside of
the sandstone–siltstone clastic rock responses. As expected, carbonaceous rocks occupy an intermediate position because they are
part coal and part clastic. On the basis of these results, regression
techniques are applied in Section 5.2 to predict sonic velocity from
the MWD data.
With the natural gamma data, the relationships with the MWD
parameters are consistent with the drilling conditions expected to
be encountered for each of the rock types present. However, a
simple trend is not evident because the natural gamma response
of sandstone is intermediate to coal and siltstone. In the case of
carbonaceous material and coal, while both of these materials
offer easy drilling conditions, carbonaceous material can have a
wide range of clay content. As a rock type, carbonaceous material
therefore ranges from being dirty coal but still with a low natural
gamma response through to carbonaceous siltstone/mudstone and
possibly tuff with a high natural gamma response.
5.2. Outlier rejection and regression
Computing the correlation coefficient often constitutes the first
step in assessing objectively any linear dependence between two
variables. To quantify the strength of possible linear relationships
between the MWD measures and geophysical attributes, Pearson
Table 3
Regression statistics. Method 1: Polynomial regression on scattered data with 15%
outlier rejection. Method 2: Fitting polynomial to piecewise continuous locally
weighted spline with regularisation. Degree k ¼7 in both cases, minimum sample
size n 41200.
Method 1
Method 2
Y
X
MSE ( 10 3)
R2
MSE ( 10 3)
R2
Bias
Velocity
Velocity
RoP
SED
7.034
7.095
0.708
0.692
5.691
4.402
0.727
0.778
2.5 10 5
3.4 10 4
Nominal drill operating conditions.
correlation coefficients were determined on the provision that
outliers were rejected. Table 2 illustrates the improvement in
correlation coefficients, as a function of increasing outlier rejection
ratio, r. With 15% outlier suppression, for example, RoP and SED
are strongly correlated with sonic velocity.
To build predictive models based on regression, two approaches were attempted. One involved fitting quadratic and
higher order polynomial regression directly to the points for various outlier rejection ratios ranging from 5% to 50%. As shown in
Fig. 6, this procedure produced unacceptable turning in the fitting
curves towards the ends of the data range.
To avoid these, a second method was developed whereby a
smoothing cubic spline16 was first used to fit local segments of the
curve using a weighted least squares procedure described in Appendix D. This provides a trade-off between smoothness and fidelity. The piecewise spline function was then approximated by a
0.1
1.6
0.09
RoP (m/s)
0.07
0.06
0.05
0.04
1.4
SED (GJ/m3 x 0.1)
Method 1
Method 2 (exact)
Method 2 (approx)
0.08
1.2
1
0.8
0.03
0.02
0.6
1.5
2
2.5
3
3.5
Velocity (km/s)
4
4.5
1.5
2
2.5
3
3.5
Velocity (km/s)
Fig. 6. Regression results. Pink curves are fitted using Method 1 which applies polynomial regression (k ¼ 7) directly to all the data samples contained in the data set with
15% outlier rejection. The resultant polynomials exhibit an unacceptable amount of undulation particularly near the end points. The blue curves are obtained using Method
2 which employs piecewise continuous locally fitted spline curves; these generally overlap almost exactly with the approximating polynomials (in the black dashed lines)
which are fitted to the splines rather than the raw data. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this
article.)
P. Hatherly et al. / International Journal of Rock Mechanics & Mining Sciences 78 (2015) 144–154
151
Table 4
Regression coefficients for predicted velocity from either RoP or SED where Y = ∑kn=0 An X n .
A7
A6
A5
A4
Y
X
A3
A2
A1
A0
Velocity
RoP
Velocity
SED
425094559
201783
17.8862
390.364
204134056
5558.724
70.59766
314.451
39394947.5
105.806
44.6989
116.3127
3866911.38
4.331583
182.919
14.3931
variation explained by the model) has increased. In addition, the
estimates are also unbiased. Table 4 provides the regression
coefficients obtained under the second method. Note, the prediction of velocity is only valid over the intervals of RoP and SED
given in Table 5.
In Fig. 7, the velocities estimated by the second method using
the RoP data in row E of blastholes (see Fig. 1) are compared with
the sonic data in those holes. This row was chosen because with
blasthole E10, there was a reasonable amount of sonic data within
the coal seam. Fig. 7 shows that for all intervals where sonic data is
available, the match between the observed and predicted sonic
velocity is good, even in coal. Furthermore, where sonic velocity is
not available, the estimations still appear to be consistent with
geological expectations.
Table 5
Regression domain.
Modelled interval [Xmin, Xmax]
Y
RoP
Velocity
0.0135
SED ( 108)
0.1045
0.4804
1.4115
7th degree polynomial. Overall, this procedure produced better
results. The mean squared error (MSE) and R2 statistics were
computed for verification purposes. From Table 3 it can be seen
the MSE has reduced under the 2nd method and R2 (the amount of
Elevation (m)
E.10
E.15
E.20
E.25
E.30
70
70
70
70
70
65
65
65
65
65
60
60
60
60
60
55
55
55
55
55
50
50
50
50
50
45
45
45
45
45
40
40
40
40
40
2
3
Velocity (km/s)
2
3
Velocity (km/s)
2
3
Velocity (km/s)
Estimated
2
3
Velocity (km/s)
2
3
Velocity (km/s)
Measured
Fig. 7. Estimated sonic velocity (black points) along row E blastholes compared to valid sonic velocity data (blue lines). The estimated points closely follow the observed
sonic data even where coal is present in the deeper sections of the holes. (For interpretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)
152
P. Hatherly et al. / International Journal of Rock Mechanics & Mining Sciences 78 (2015) 144–154
south
north
Fig. 8. Interpolation of blasthole data along row B blastholes. (Top) natural gamma data, (middle) RoP and (bottom) SED. Within each section, the white lines indicate the top
of the upper carbonaceous layer, the top and bottom of the main coal seam and the top of the lower seam. The interpolations between the blastholes were guided by these
boundaries. The natural gamma results show the presence of sandstone (green) and siltstones (yellow) above the coal seams (blue). The MWD (RoP and SED) data clearly
match the stratigraphy shown by the natural gamma results. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of
this article.)
It is stressed that the regressions performed here are only relevant to the geology, drills and drill operating parameters encountered in this study. Prediction of velocity from other MWD
data should not be attempted using the derived coefficients of
Table 4. Nevertheless, the general principles that have been demonstrated remain applicable for other studies.
These results show that because MWD data can be used to
predict sonic velocity, it could also be used to empirically estimate
rock strength in the same way that sonic velocity is used as a
proxy for rock strength and quality. Published applications for
sonic velocity, include rippability assessments,10,18 unconfined
compressive strength (UCS) estimation7,11 and estimation of Q-
value.12 It also forms the basis of the geophysical strata rating
(GSR).8 In the context of blasting studies, the widely used Kuz–
Ram model for estimating blast fragmentation utilises a hardness
factor based on UCS.17 Clearly, MWD data have the potential to be
used in blast design.
6. Spatial modelling between blastholes
To further demonstrate the relationship between the bench
geology revealed by the geophysical logging data and the MWD
results, Fig. 8 shows examples of 2D interpolation (modelling) of
P. Hatherly et al. / International Journal of Rock Mechanics & Mining Sciences 78 (2015) 144–154
parameters between blastholes along row B of blastholes. The
interpolation method used here involves point by point linear
interpolation of geophysical and MWD parameters between holes,
where the interpolations are conducted between points at levels
established on the basis of positions with respect to the boundaries of the geological layers. These boundaries were established
by initial interpolations (using splines) between the depths to the
layers within the blastholes (see description in Ref. 9). The layers
used to guide the interpolation were the top of the top carbonaceous band, the top of the main seam, the base of the seam and
the top of the underlying seam.
The results of the interpolation of the geophysical data
shown in Fig. 8 are for the natural gamma logs. As previously
discussed, the model confirms that above the coal seams,
sandstones are more prevalent to the south (left of model) and
that they finger out and grade into siltstones to the north.
Within the coal, the mid-seam band is clearly evident and thins
to the north. The MWD results (RoP and SED), show the same
patterns of behaviour as the natural gamma model and the same
major features. For example, the zones of clean sandstone in the
natural gamma model between 58–60 m and 54–55 m in blasthole B20 are matched by corresponding changes in the RoP and
SED results.
Comparison of the models also suggests that the MWD results
have reduced vertical resolution. The natural gamma model shows
structure at the sub metre vertical scale that is correlated between
holes and represents true geological banding. In the case of the
MWD models, there are some features at the sub metre scale that
correlate between holes, but there are not as many of these as in
the natural gamma model. This may be indicating that the MWD
data are being smoothed and filtered by the MWD system on the
drills (sensors and/or recording system), the details of which were
not available. Notwithstanding, the current results show that the
MWD data preserve the general geological variability revealed by
the geophysical logs and can be used for detailed geotechnical
analysis.
7. Conclusions
This study has provided the opportunity to make a detailed
assessment of MWD data from blasthole drills with the geological
conditions revealed by geophysical logs obtained in the same
holes. For both the MWD data and the geophysical logs, careful
treatment and filtering of the data were required to properly
correlate the measurements in depth and understand the data.
Once this was done, it was found that there is a very good correlation between the MWD measures (RoP and SED) and rock type
and also rock strength as revealed by sonic velocity. It has been
demonstrated that RoP and SED can be used to predict sonic velocity. This means that with proper calibration and drilling under
nominal conditions where the rotary speed and WoB remain
constant as far as possible, MWD measurements can be used to
determine the rock properties needed for blast design using approaches such as the Kuz–Ram fragmentation model. It is demonstrated that interpolation of MWD data between blastholes
provides a similar view on changing geological conditions to that
provided by geophysical logs.
153
Innovation (T&I), Coal and Allied (RTCA) and HVO personnel for
providing the drill data used in the experiments. Charles McHugh
is also thanked for his assistance in setting up the trials and encouraging studies such as this.
Appendices
These appendices embody the more significant computational
aspects of this work and present solutions to practical problems
encountered during data processing. The techniques employed are
also applicable to analysis and modelling of scattered data in a
wider context.
Appendix A. Outlier removal in MWD data
In addition to the method described in Section 4 for replicating
data over intervals where rod changes occurred, the MWD data
were also scanned for other outliers where any of the following
conditions were met. These data were removed from the analysis.
RoP < μ rop − 2σrop
Torque ≥ μ torque + 2σtorque
WoB ∉ [μ wob − σwob, μ wob + σwob ]
Rotaryspeed ∉ [μ rotary − σrotary, μ rotary + σrotary ]
Appendix B. Sequence alignment
Since the geophysical data and MWD measurements were acquired through two separate processes, these signals first needed
to be brought into alignment with one another in a common coordinate frame, before they could be integrated. As also explained
in the text, a relative offset exists between the geophysical and
MWD data, which varies between blastholes.
The technique we employed to estimate the offsets uses local
feature based cross-correlation. This was implemented using the
fast Fourier transform and peak detection, and the results were
confirmed by visual inspection. The key here is ‘local’ as opposed
to ‘global’ sequence correlation. To obtain sensible estimates,
relevant features needed to be accentuated with weights (and the
irrelevant parts attenuated accordingly) to minimise distraction
during the comparison. Fig. 9 illustrates this localised pattern
matching idea and the shift required to maximise the alignment
between bulk density (from geophysics) and the APR sequence (a
MWD measure proportional to the RoP19) for a typical blasthole.
Statistically, the offset is roughly Gaussian distributed with a
mean of 0.75 m, median of 0.70 m and standard deviation of
0.29 m. A positive value is interpreted as requiring the origin of
MWD depth to be shifted down (for that hole) in order to match
the true ground surface. After noise filtering and offset compensation, the geophysics data were then down-sampled to
produce sequences with resolution matching those of the MWD
measurements.
Appendix C. Heat maps via Kernel Density Estimation
Acknowledgements
This work has been supported by the Australian Centre for Field
Robotics and the Rio Tinto Centre for Mine Automation. The authors would like to acknowledge Rio Tinto Technology &
To develop a firm grasp of the relationships observed in the
scatter plots, the conditional distribution of points with respect
to the X–Y variables is considered. The motivation is to understand where each data cluster is concentrated, or is most likely to
154
P. Hatherly et al. / International Journal of Rock Mechanics & Mining Sciences 78 (2015) 144–154
the scatter data to ensure the curve conforms better to the expected trend.
Method 2 uses a smoothing cubic spline16 to fit local segments
of the curve, and provides a trade-off between smoothness and
fidelity. Mathematically, the optimisation problem is formulated as
N
argminθ p ∑ wi |yi − s (xi )|2 + (1 − p)
i=1
0
50
APR sequence
100
150
Density
200
Features of interest
250
300
Matched peaks
Fig. 9. Essence of weighted cross-correlation based on local feature matching. This
sequence alignment technique is applied to compensate for the offset that exists
between MWD and geophysical measurements. After reconciliation, both sequences are expressed relative to a common frame.
appear, in the X–Y plane. An obvious approach is to build a 2D
frequency histogram. Unfortunately, this approach is biased toward dense clusters, as it allows dominant point clusters to
overshadow other points which appear more sparsely along
major trend lines. To make the contribution of each group of
points more equitable, data points are partitioned into subsets
(different classes) that constitute representative clusters. In this
work, a class corresponds implicitly to a rock type. This is done by
labelling samples from the 35 blastholes as either “sandstone”,
“siltstone”, “carbonaceous” or “coal” using density and gamma
logs as described in Section 3. In situations where such labels are
unavailable, alternatives such as the K-mean algorithm may be
used to assign points to clusters (classes) in an unsupervised
manner.
To capture the X–Y relation across the entire domain, the likelihood p(x,y|c) is computed for each class c using the Kernel Density
Estimation method proposed in.20 The ultimate goal is to produce
heat maps which indicate where in the X–Y plane the points are
most likely to fall regardless of the prior probability (i.e., ignoring
the abundance or scarcity of samples from a particular class). The
heat maps shown in Fig. 5, are essentially blended images of these
likelihood functions over all classes (rock types).
The heat maps are generated by taking the maximum likelihood
amongst all classes at each location. Formally, this may be expressed as fheat (x, y ) = supc p (x, y|c ) where “sup” denotes the supremum. This kind of representation is of general interest when one
wishes to highlight the degree of correlation between two variables
irrespective of class, or emphasise the separation between class
clusters. The heat maps are useful for a number of processing tasks,
including outlier rejection and weighted regression.
Appendix D. Robust regression
The main deficiencies with applying standard polynomial regression to the scattered data (Method 1) is the undulation observed in the resultant polynomial, especially near the end points.
The turning and twisting behaviour may be attributed to inadequate suppression of peripheral data points and competing
forces causing the curve to bend in order to satisfy global least
squares constraints with a low degree of freedom.
In relation to the first point, inadequate suppression of distant
points (from the apparent trend) is due to Method 1 treating all
points in the data set as equally important. Points that are spread
out, or located at the extremities, currently contribute more than
their fair share to the curve fitting error. The proposed solution to
this is to attach an importance weight to each point in such a way
as to emphasise centrality.
In regard to the second point, to minimise undulation, we
propose fitting a piecewise continuous curve to local sections of
∫ λ |s″(t; θ )|2dt
(1)
where s (x i ) denotes a local spline interpolation and p ∈ [0, 1] is a
regularisation parameter. The first term represents a weighted
sum of spline fitting errors and the second term represents a
penalty for over-fitting which in practice is based on the second
derivative of the continuous spline, expressed as s (t ) in parametric
form. For more details, see Ref. 21.
To leverage our KDE (kernel density) computation, we set the
weights equal to the likelihood estimates depicted in the heat
maps. Finally, as the piecewise continuous spline yields a rather
dense representation, it is subsequently approximated using a
polynomial of degree k. Empirically, a value of k ¼7 was determined as sufficient for minimising the spline approximation
error.
References
1. McHugh C, Stokes A, Oppolzer F, Rogers B. Automated multi-drills in Rio Tinto.
In: Proceedings of the Eighth Open Pit Operators' Conference. AusIMM.
Melbourne; 2012:83–88.
2. Scoble M, Peck J. A technique for ground characterisation using automated
production drill monitoring. Int J Min Reclam Environ 1987;1:41–54.
3. Schunnesson H. Rock characterisation using percussive drilling. Int J Rock Mech
Min Sci 1998;35:711–25.
4. Finfinger GL, Wilson G, Peng S, Thomas B, Gu Q. An approach to identifying
geological properties from roof bolter drilling parameters. In: Proceedings of
the 19th International Conference on Ground Control in Mining; 2000:1–11.
5. Teale R. The concept of specific energy in rock drilling. Int J Rock Mech Min Sci
Geomech Abstr 1965;2:57–73.
6. Leung R, Scheding S. Automated coal seam detection using a modulated specific
energy measure in a monitor-while-drilling context. Int J Rock Mech Min Sci
2015;75:196–209.
7. McNally G. The prediction of geotechnical rock properties from sonic and
neutron logs. Explor Geophys 1990;21:65–71.
8. Hatherly PJ, Medhurst TP, MacGregor SA. A rock mass rating scheme for clastic
sediments based on geophysical logs. In: Proceedings of the International
Workshop on Rock Mass Classification in Underground Mining, Canada–USA
symposium on rock mechanics. Vancouver; 2007:57–63.
9. Hatherly PJ, Medhurst TP, Zhou B. Geotechnical modelling based on geophysical
logging data. In: Proceedings of the13th Coal Operators' Conference; 2013:
21–26.
10. Caterpillar Inc. Caterpillar Performance Handbook. 38. Peoria, Illinois: Caterpillar Inc.; 2008.
11. Oyler DC, Mark C, Molinda GM. In situ estimation of roof rock strength using
sonic logging. Int J Coal Geol 2010;83:484–90.
12. Barton N. Some new Q-value correlations to assist in site characterization and
tunnel design. Int J Rock Mech Min Sci 2002;39:185–216.
13. Scoble M, Peck J, Hendricks C. Correlation between rotary drill performance
parameters and borehole geophysical logging. Mining Sci Technol 1989;8:
301–12.
14. Martin Gonzalez J. Application of Pattern Recognition Techniques to Monitoring-While-Drilling on a Rotary Electric Blasthole Drill at an Open Pit Coal Mine
(M. Eng thesis). Kingston Ontario: Queen's University; 2007.
15. Sniffin M, Beckett J. Sydney basin Hunter coalfield. In: Geol Australian Coal
Basins. Geol. Soc. Aust. Coal Geol. Group; 1995:177–195.
16. Boor C de. A Practical Guide to Splines, Applied Mathematical Sciences, vol. 27.
Berlin: Springer-Verlag; 2001.
17. Cunningham C. The Kuz–Ram fragmentation model – 20 years on. In: Proceedings of 3rd EFEE World Conference on Explosives and Blasting. England;
2005:201–210.
18. MacGregor F, Fell R, Mostyn G, Hocking G, McNally G. The estimation of rock
rippability. Q J Eng Geol 1994;27:123–44.
19. Zhou H, Hatherly PJ, Monteiro ST, et al. Automatic rock recognition from drilling
performance data. In: Proceedings of the IEEE International Conference on
Robotics and Automation; 2012:3407–12.
20. Botev Z, Grotowski J, Kroese D. Kernel density estimation via diffusion. Ann Stat
2010;38:2916–57.
21. Barry J. Splins and SMOOTH: Two Fortran smoothing routines. Tech Rep AAECE-253, ANSTO.1973. URL 〈http://apo.ansto.gov.au/dspace/handle/10238/444〉.