Improvements in Blast Fragmentation Models Using Digital

Image Processing
J Kemeny1, E Mofya1, R Kaunda1 and P Lever2

ABSTRACT is used to assess post-blast fragmentation and can be correlated

One of the fundamental requirements for being able to optimise blasting
with rock mass and blasting information on a hole by hole basis.
is the ability to predict fragmentation. An accurate blast fragmentation In addition, the primary crusher feed and product size
model allows a mine to adjust the fragmentation size for different distributions are used to assess the work index of the fragmented
downstream processes (mill processing versus leach, for instance), and to rock. This is important, since blasting can also improve the
make real time adjustments in blasting parameters to account for changes crushability and grindability of individual fragments of ore due
in rock mass characteristics (hardness, fracture density, fracture to the introduction of micro-fractures (Nielsen and Malvik,
orientation, etc). A number of blast fragmentation models have been
1999). Finally, on the blast-hole drills, drill monitoring systems
developed in the past 40 years such as the Kuz-Ram model
(Cunningham, 1983). Fragmentation models have a limited usefulness at are installed. The drill monitoring information is used to assess
the present time because: the pre-blast conditions of the rock mass on a hole-by-hole basis.
1. The input parameters are not the most useful for the engineer to
determine and data for these parameters are not available THE SPLIT SYSTEM
throughout the rock mass.
An effective method to assess fragmentation at the present time
2. Even if the input parameters are known, the models still do not is to acquire digital images of rock fragments and to process
consistently predict the correct fragmentation. This is because the
models capture some but not all of the important rock and blast these images using digital image processing techniques. In the
phenomena. case of post-blast fragmentation, this is the only practical method
to estimate fragmentation, since screening is impractical on a
3. The models do not allow for ‘tuning’ at a specific mine site.
large scale. In the case of post-crusher fragmentation, screening
This paper describes studies that are being conducted to improve blast
fragmentation models. The Split image processing software is used for
is routinely used, but digital image processing allows
these studies (Kemeny, 1994; Kemeny et al, 1999). fragmentation to be assessed on a continuous basis. The
development of image processing techniques for the assessment
of fragmentation has been in development at the University of
INTRODUCTION Arizona from 1990 through 1997. Since 1997 the development
The Split software was originally developed at the University of work has continued at Split Engineering, LLC, and professional
Arizona, and in 1997 the technology was transferred to a newly Split-Online systems have now been installed at over 38
formed company, Split Engineering. The Split software allows locations worldwide.
post-blast fragmentation to be determined on a regular basis The basic steps involved in the Split software are as follows:
throughout a mine, by capturing images of fragmented rock in
muckpiles, on haul trucks, or from primary crusher feed or 1. acquiring digital images, either automatically or manually;
product. The resulting size distribution data can then be used to 2. pre-processing the images to correct for lighting problems
accurately assess the fragmentation associated with different and to screen for unacceptable images;
parts of a shot. And in particular, this data can be used to assess
and improve the accuracy of fragmentation models (Higgins et 3. delineate the individual fragments in each image using
al, 1999). digital image processing algorithms;
Fragmentation models are also being improved by utilising 4. apply statistical algorithms to the 2D particle areas in each
drill-monitoring data. Drill-monitoring data includes raw drilling image to determine 3D particle volumes;
data such as rotary torque, penetration rate, and pull down 5. statistically correct 3D volumes for overlap and shape and
pressure, as well as calculated quantities such as drilling specific determine histogram of particle volumes;
energy or the Aquila Blastability Index (Peck and Gray, 1995).
Because drill-monitoring data is available from every blast hole, 6. correct particle volume histogram for fines;
it provides data throughout the rock mass to be blasted. 7. process multiple images together to get an average
As part of this project fragmentation studies are being distribution (including images taken at different scales); and
conducted at several large open pit mines in Arizona. At these
8. output data to the screen, hard disk, and network control
mines Split-Online systems are installed at the primary crushers.
systems.
On these systems, cameras installed at the truck dumps monitor
primary crusher feed and cameras installed at the discharge belts Details of these steps are given in Kemeny et al (1999) and
monitor primary crusher product. The primary crusher feed Kemeny (1994). Figure 1a shows a typical image of primary
information is then traced back to the original position of this crusher product and Figure 1b shows the delineation of the
rock on the shot using mining dispatch systems. This information individual rock fragments by the Split software.
At typical open pit mining applications, a Split-Online system
is installed at the primary crusher, where digital images of both
1. Department of Mining and Geological Engineering, University of feed and product are continually processed and the results are
Arizona, Tucson Arizona 85721, USA.
recorded. The feed cameras are located at the truck dump bays or
2. Chief Operating Officer, Deputy CEO and Mining Systems Program feed belts, and the product cameras are located above the
Leader, Cooperative Research Centre for Mining Technology and discharge belts. The resulting size data from the Split system can
Equipment (CMTE), Isles Road, Indooroopilly Qld 4068. E-mail: be imported into a mine-wide database where truck-by-truck
p.lever@cmte.org.au averages of the feed and product sizes can be determined.

EXPLO 2001 Hunter Valley, NSW, 28 - 31 October 2001 213

J KEMENY, E MOFYA, R KAUNDA and P LEVER

A Split-Online system installed at a primary crusher serves

four functions. First of all, the crusher feed size provides
information on post-blast fragmentation. Secondly, the crusher
product size provides information on secondary crusher and ball
mill feed size. Thirdly, the feed and product sizes together can be
used to estimate the work index, which gives information on the
crushability and grindability of the ore. And finally, the feed and
product sizes can be used to monitor crusher performance and
crusher wear.
Several new technologies are being utilised to trace the crusher
feed and product size information back to the original position of
the rock on the bench. This is accomplished on a truck-by-truck
basis. The technologies utilised include an accurate time/date
stamp for the Split data associated with each truckload of ore,
mining dispatch systems to trace the trucks back to the bench,
and GPS equipped shovels to determine the locations of the
material dumped into each truck. Figure 2 shows an example of a
bench along with the locations of Split crusher data. The F80 A
values around each hole are averaged, and this hole-by-hole data
is used in the development of fragmentation models, as described
in Section 4.
Bond’s equation is used to estimate the work index of the
material going through the primary crusher (Bond, 1952):
P
W= = 10Wi ( P80− 0 .5 − F80− 0 .5 )
T
where
W is the work input in kWh/t
Wi is the work index of the ore in kWh/t
P80 is the 80 per cent passing size of the product
F80 is the 80 per cent passing size of the feed
T
P
is the throughput of new feed in t/h
is the power draw in kW
B
Bond’s equation can be rearranged to estimate the work index:
P
Wi = − 0 .5
10 T( P 80 − F80− 0 .5 ) FIG 1 - a) Image of primary crusher product. b) Delineated image using
the Split software. Scale in the image is six inches.
Thus by knowing the feed and product sizes and the power
draw and throughput, the work index for the material passing
Drill Holes
through the crusher can be estimated. Split F80

100150
DRILL MONITORING DATA
100100
One of the factors that has limited the usefulness of blast
fragmentation models is the lack of information on the in situ 100050

characteristics of the rock mass. This information includes rock

Y- Coordinates

100000
strength, RQD, fracture spacing, and fracture orientation. In most
mining environments these parameters will be highly variable 99950
even within a single shot, and the traditional methods for
obtaining this information, diamond drill core and geologic 99900

mapping, cannot collect the quantity of information necessary for

99850
fragmentation models. A new approach to obtaining at least
some of the rock mass properties is the use of drill monitoring 99800
data. This information is available prior to blasting and 96950 97000 97050 97100 97150 97200
throughout the rock mass to be blasted. Typical parameters X - Coordinates
recorded by a drill monitoring system include depth (ft),
penetration rate (ft/hr), pull-down force (lbs), rotary speed (rpm),
bit air pressure (psi), rotary current (amps) and rotary torque FIG 2 - A blasting shot showing the locations of drill holes and also F80
(lb-ft). fragmentation size information. The F80 information is from a
Drilling parameters have been used in a number of studies to Split-online system installed at the primary crusher.
estimate rock drillability or blastability (Protodyakonov, 1962;
Peshalov, 1973; Miller, 1972; Schmidt, 1972; and Rabia, 1980).
The most common approach to predicting drillability and Consider the following drilling parameters: penetration rate
blastability is the concept of specific energy. Teale (1965) defines (PR, in/min), torque (T, lb-in), rotational speed (N, rev/min),
specific energy as the work done per unit volume of rock cross section area of drill hole (A, in2), and pulldown force
excavated. (F, lb). The work done per minute is given by W = F PR + 2 π N

214 Hunter Valley, NSW, 28 - 31 October 2001 EXPLO 2001

 IMPROVEMENTS IN BLAST FRAGMENTATION MODELS USING DIGITAL IMAGE PROCESSING

T. The volume of material excavated in a minute is given by V = allowed important pre and post-blast information to be obtained
A PR. The specific energy SE is then given by: on a hole-by-hole basis. This includes the Split size information
w F PR + 2π NT F 2π NT and the drilling specific energy (SE). In addition, the explosive
SE = = = + energy per ton of rock (kcal/t) can also be estimated on a
v A PR A A PR hole-by-hole basis. At the present time the focus of the
modelling is on the following five quantities that can be obtained
Specific energy can be thought of as having two components,
on a hole-by-hole basis:
one due to the pulldown force and another due to the torque.
Previous studies have shown that the component of specific 1. SE (drilling energy),
energy from the pulldown force is very small compared to that 2. kcal/t (explosive energy per volume),
from torque, typically less than five per cent (Teale, 1965;
Schivley, 1994; Karanam and Misra, 1998). For all practical 3. F80 (post-blast 80 per cent passing size),
purposes, the first term in the equation above is negligible and 4. P80 (post primary crush 80 per cent passing size), and
can be dropped out, leading to the equation below.
5. Wi (work index).
2π NT
SE = These quantities take into account the in situ characteristics of
A PR
the rock mass, the blasting parameters and the resulting
Typical variations of specific energy with depth are shown in fragmentation size and strength. One approach to analysing this
Figure 3 for three adjacent drill holes. This data was acquired data is to determine parameters needed for existing
during a normal mine production blasthole drilling operation at fragmentation models from the obtained information. For
an open pit mine in Arizona. example, the uniaxial compressive strength can be estimated
from the drilling specific energy. The approach taken here,
however, is to develop fragmentation models that take as their
input (and output) the specific quantities given above.
40 Using data from standard and experimental blasts at a mine in
35 Arizona, statistical relationships between SE, F80, P80, Wi, and
kcal/t are being investigated. Some sample results from several
30
SE (psi x 1000)

shots at a mine in Arizona are given below. Figures 4a and 4b

25 present results from two blasts, a high-energy blast (average 250
kcal/t) and a low energy blast (average 150 kcal/t). Each of the
20 points in the figures represents the data from an individual blast
hole. The filled squares are data points from the low energy blast,
15 and the unfilled squares are data points from the high-energy
10 blast.
Figure 4a is a plot of F80 as a function of drilling SE. The F80
5 is representative of the post-blast fragmentation and SE is
0 representative of the strength of the pre-blast rock mass. First of
Depth (ft) all, on average Figure 4a shows a significant decrease in
0 20 40 60 fragment size with increasing blasting energy. Secondly, the
figure shows that for the low energy blast the fragmentation size
FIG 3 - Drill specific energy (SE) versus depth for three adjacent blast is somewhat sensitive to the hardness of the in situ rock (SE), but
holes. The top and middle curves are offset by 20 000 and 10 000 psi, for the high energy blast a correlation between F80 and SE is not
respectively, for better clarity. evident. Figure 4b is a plot of F80 as a function of explosive
energy per ton. This plot clearly separates the high and
low-energy shots into two groups, with the high energy shot
From Figure 3 it can be seen that the specific energy has a showing a significant reduction in F80 on average. It is worth
gentle upward trend with depth. This is expected due to: noting that Figure 4b shows the large variation in hole-to-hole
• increasing rock hardness with depth due to confining blasting energy for each of the shots.
pressure; and Overall, the main conclusions from Figures 4a and 4b involve
the large decrease in F80 with increasing explosive energy. More
• drilling through and below the damage zone created from the subtle conclusions about specific relationships on a hole-by-hole
previous blast.
basis between SE, F80 and KCal/ton are difficult to make at the
For the purpose of correlating specific energy with blast present time due to the small sample size and the large scatter in
fragmentation, a single specific energy value was obtained for the data. Much of the scatter is probably due to heterogeneities in
each hole by averaging all the interval specific energies of each the rock mass properties. Some of the scatter may also be due to
hole. errors, such as the error in estimating the exact location of the
In addition to the gradual upward trends, the curves in Figure 3 bucket when it digs and the error in assigning this ore to a
show sudden changes at specific depths due to lithologic changes specific drill hole due to throw and mixing. As more data is
or fractures. Some of these changes occur at similar depths in collected at the mine and the overall sample size becomes much
adjacent holes. Current studies are being conducted to identify bigger, statistically significant relationships should become
the nature and properties of the discontinuities based on drill apparent, and these relationships will form the basis for a
monitoring data such as the data shown in Figure 3. mine-specific fragmentation model.

THE DEVELOPMENT OF FRAGMENTATION CONCLUSIONS AND FUTURE WORK

MODELS This paper presented an approach for developing accurate models
An approach has been developed for the optimisation of blasting to predict the fragmentation due to blasting. The approach makes
and the development of accurate blast fragmentation models. As use of drill monitoring data to provide information on the in situ
described in the previous sections, recent technologies have rock mass, Split image processing software for assessing

EXPLO 2001 Hunter Valley, NSW, 28 - 31 October 2001 215

J KEMENY, E MOFYA, R KAUNDA and P LEVER

This work is ongoing, and several additional aspects may be

investigated in the future. Most importantly, data from the mines
14 using this technique is continually being collected, resulting in
improved correlations with time. Also, in addition to the F80 and
P80 size information, additional size information can be used in
12 the statistical analysis, including the F20, P20, F50 and P50.
Additional blasting information could be utilised in the future,
including the explosive geometry, particle velocity and
detonation timing. Additional information about the rock mass
F80 (inches)

10
could also be collected, including digital images of rock faces for
determining detailed fracture information. Finally, laboratory
8 tests are being planned to investigate changes in mechanical
properties that occur due to changes in in situ and blasting
conditions.
6

Low KCal/ton ACKNOWLEDGEMENTS

4 High KCal/ton
4000 6000 8000 10000 12000 14000
This work is being funded under DOE Industries of the
Future/Lawrence Berkeley Laboratory contract 6496612.
SE (psi)
REFERENCES
14 Bond, F C, 1952. The third theory of comminution, Transactions AIME,
193, 484.
Cunningham, C V B, 1987. Fragmentation estimation and the Kuz-Ram
12 model – Four years on, in Proceedings Fragblast ’87, Keystone,
Colorado, pp 475-487.
Higgins, M, Seppala, V, Kemeny, J, BoBo, T and Girdner, K, 1999.
Integrated software tools and methodology for optimization of blast
F80 (inches)

10
fragmentation, Proceedings of the International Society of Explosive
Engineers annual meeting, Nashville, TN.
8 Karanam, U M R and Misra, B. Principles of Rock Drilling, pp 111 – 144
(A A Balkema).
Kemeny, J, 1994. A practical technique for determining the size
6 distribution of blasted benches, waste dumps and heap-leach sites,
Low KCal/ton
Mining Engineering, 46(11):1281-1284.
Kemeny, J, Girdner, K, BoBo, T and Norton, B, 1999. Improvements for
4 High KCal/ton fragmentation measurement by digital imaging: accurate estimation
100 200 300 of fines, in Proceedings FragBlast 6, (South African Institute of
Mining and Metallurgy: Johannesburg).
Explosive Energy (kcal/ton)
Miller, M. Normalization of specific energy, Int J Rock Mech Min Sci, pp
661-663.
Nielsen, K and Malvik, T, 1999. Grindability enhancement by
blast-induced microcracks, Powder Technology, 105:52-56.
FIG 4 - a) Drilling specific energy (SE) versus primary crusher F80.
Peck, J and Gray, J, 1995. Total Mining Systems (TMS): the basis for
Results from two blasts, a high-energy blast (average 250 kcal/t, unfilled
open-pit automation, CIM Bulletin, 88(993):38-44.
squares) and a low energy blast (average 150 kcal/t, filled squares).
Peshalov, Y A. Resistance of rocks to breaking by impact loads due to
b) Explosive Energy (kcal/ton) versus primary crusher F80. Each of the
drilling, Soviet Mining Science, Vol 9, pp 97-100.
points in the figures represents the data from an individual blast hole.
Protodyakonov, M M, 1962. Mechanical properties and drillability of
rocks, in Proceedings Fifth Symposium on Rock Mechanics, pp
103-118.
post-blast fragmentation and the crushability and grindability of
Rabia, H and Brook, W, 1980. An empirical equations for drilling
the ore, and the explosive energy per unit volume of rock. These performance prediction, in Proceedings 21st US Symposium on Rock
three types of data are collected and analysed on a hole-by-hole Mechanics, pp 103-111.
basis, giving 50 or more data points for each blast. These data Schivley, G P. Predicting rotary drill performance, 20th Annual
points form the basis for a statistical correlation between in situ Conference on Explosive and Blasting Techniques.
conditions, blasting parameters, and the resulting fragmentation Schmidt, R L, 1972. Drillability studies – Percussive drilling in the field.
size and strength. At a specific mine, the database is continually USBM RI 7684.
updated as mining progresses, resulting in an evolving and Teale, R, 1965. The concept of specific energy in rock drilling, Int J Rock
increasingly accurate model with time. Neural networks or other Mech Min Sci, pp 57-73.
learning algorithms are well suited for handling this evolving Tunstall, A and Bearman, R, 1997. Influence of fragmentation on
fragmentation model, and these types of models will be crushing performance, Mining Engineering, Vol 49, pp 65-70.
investigated in the future. Some sample results from a mine in Wills, B A, 1997. Mineral Processing Technology, Sixth Edition
Arizona have been presented. Even with the small data set (Butterworth-Heinemann: Oxford).
shown, these results show the potential usefulness of being able
to analyse data on a hole-by-hole basis.

216 Hunter Valley, NSW, 28 - 31 October 2001 EXPLO 2001