Você está na página 1de 12

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/307594360

Comparative Analysis of Rock Fragmentation Models – A Case Study

Conference Paper · August 2016

CITATION READS

1 964

3 authors:

Bruno Ayaga Kansake Victor Temeng


Missouri University of Science and Technology University of Mines and Technology
10 PUBLICATIONS   4 CITATIONS    14 PUBLICATIONS   66 CITATIONS   

SEE PROFILE SEE PROFILE

Bright Afum
Laurentian University
7 PUBLICATIONS   5 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Mercury Pollution in a Fresh water body in Ghana View project

Preparedness of Ghanaian mine stakeholders for the adoption of autonomous mining systems View project

All content following this page was uploaded by Bruno Ayaga Kansake on 03 September 2016.

The user has requested enhancement of the downloaded file.


Comparative Analysis of Rock Fragmentation Models – A Case
Study
B. A. Kansake, V. A. Temeng, B. O. Afum

Kansake, B. A., Temeng, V. A. and Afum, B. O. (2016), “Comparative Analysis of Rock Fragmentation Models
– A Case Study”, 4th UMaT Biennial International Mining and Mineral Conference, pp. MP 1 – 11.

Abstract
The products of blasting affect all downstream operations of mining such as loading, hauling and processing. Therefore, blasts
should be designed to ensure that selected parameters give the desired fragmentation to optimise downstream processes. This
may be achieved by using fragmentation prediction models. Several models exist for predicting fragmentation distribution
from primary rock blasting. The Kuz-Ram model is the most widely used fragmentation prediction model. This research
assesses the performance of the Kuz-Ram, Modified Kuz-Ram and KCO models to determine the most accurate model
applicable for the mines where the study was conducted. Data were obtained from two Ghanaian mines: Mine A and Mine B.
The performance assessment was done using the Root Mean Square Error and correlation and regression analysis. A general
trend of over predicting fines (< 15.24 cm) and boulders (> 100 cm) was observed for the blasts at Mine A while a trend of
underestimation of the fines (< 10.16 cm) and boulders (> 80 cm) was observed at Mine B using all the three models. The
fragmentation analysis results reveal that there was high quantity of fines with insignificant amount of boulders produced from
all the blasts studied. Though all the models had high correlation coefficients, R (above 95%), the Modified Kuz-Ram model
performed best at Mine A while the Kuz-Ram model performed best at Mine B. They are therefore recommended for
fragmentation prediction and blast optimisation studies at the respective mines. The KCO and Modified Kuz-Ram models did
not perform better than the Kuz-Ram model in predicting fines as stated by earlier researches.

Key words: Comparative analysis, Fragmentation Prediction, Models, Kuz-Ram.

1 Introduction Cho and Kaneko, 2004; Johnson, 2014). These


models range from purely empirical relations to
Blasting is the first step in the rock comminution rigorous numerical models (Kanchibotla et al.,
process, and is the most efficient fragmentation 2001). Some of these models pertaining to surface
method in hard rock mining (Vasileois, 2008; blasting include the Bond-Ram model, Kou-Rustan
Siddiqui et al., 2009; Engin, 2010). Afum (2014) equation, Energy Block Transition (EBT) model,
defines blasting as a process that involves placing Swedish Detonic Research Foundation (SveDeFO)
chemical substances called explosives into model, Kuz-Ram model, Larson model, Rosin-
blast/drill holes and detonating them to cause an Rammler model, Kuznetsov–Cunningham–
explosion leading to the breaking of the rock Ouchterlony (KCO) model, Chung and Katsabanis
formation into smaller fragments for an appropriate model, Modified Kuz-Ram model, Crushed Zone
end use. Model (CZM), Two Component Model (TCM),
Artificial Neural Networks (ANN) and many more
In mining, the rock fragments obtained from (Kanchibotla et al., 2001; Ouchterlony, 2005;
blasting are usually further processed to liberate the Kazem and Bahareh, 2009; Vamshidhar and
minerals of interest or to attain the appropriate end Venkatesh, 2010).
use fragment sizes. The characteristics of these
fragments are very important as they affect the These models consider a wide range of factors
overall efficiency and cost of downstream processes (controllable and uncontrollable) such as
including loading, hauling and processing (Siddiqui geometrical (burden, spacing, hole depth and charge
et al., 2009). It is therefore imperative that much length), explosive (type, amount/quantity and
attention is given to blast design. This can be done properties) and rock (rock strength, porosity,
by using the available fragmentation prediction specific gravity, discontinuity information and
models to predict the fragmentation distribution that ground water condition) parameters to predict the
results from using a particular set of blasting fragmentation. However, the commonly used model
parameters and hence selecting those that give the for surface fragmentation prediction is the Kuz-Ram
optimum/required fragmentation. model due to the ease with which it can be
parameterised and its direct linkage between blast
Various models have been developed over the years design and rock breakage results (Cunningham,
that predict the size distribution resulting from a 2005).
particular primary blast design (Cunningham, 2005;

MP 1
Though the Kuz-Ram model (Johnson, 2014; Kuznetsov’s equation
Cunningham, 2005; Djordjevic, 1999) is the most
widely used empirical fragmentation prediction The Kuznetsov equation (Eqn. 1) is used in
model, it has several drawbacks which subsequent estimating the mean fragment size resulting from a
models sought to address. A major weakness is that, blast.
it underestimates the quantity of fines produced
from a blast. To address this weakness, the 19
 0.8
1 115  20
uniformity index used in the Rosin-Rammler X m  AK Q  6
 (1)
equation (Eqn. 2) was modified using  RWS 
Cunningham’s uniformity index (Eqn. 3) and
blastability index (Eqn. 10) to obtain Eqn. 9. This, where Xm is the mean particle size, cm; A is rock
together with a modification in the Kuznetsov factor (usually between 0.8 and 22 depending on the
equation (Eqn. 1), resulted in the modified Kuz-Ram nature of the rock); K is the powder factor, kg/m3; Q
model (Gheibie et al., 2009; Vamshidhar and is the quantity of explosives per hole, kg; RWS is
Venkatesh, 2010). Also, the Swebrec function (Eqn. the relative weight strength of the explosive used
6) was introduced to replace the Rosin-Rammler (RWS of ANFO = 100); and 115 is the relative
equation in an effort to improve fines prediction. weight strength of TNT.
The uniformity index was also replaced by the curve
undulation parameter (Eqn. 7). These modifications Rosin-Rammler equation
also resulted in the KCO model (Vasileois, 2008).
This is used for predicting the distribution resulting
Another weakness of the Kuz-Ram model is that it from the blast. It is as given in Eqn. 2.
does not consider accurate timing or delay offered
by modern electronic detonators. It also assumes   X  
n
(2)
R x  exp   0.693  
that the energy released by explosives in adjacent
  Xm  
holes does not interfere with each other. This cannot
be true with precise shorter delays offered by where Rx is the mass fraction retained on screen
electronic/electric blasting. opening X and n is the uniformity index usually
between 0.7 and 2 based on the blast geometry.
Therefore, the need to make a comparative study of
these models to ascertain their accuracy and whether Uniformity index equation
the modifications made to the Kuz-Ram model have
improved fines prediction by comparing prediction This equation (Eqn. 3) is used to characterise the
results with field results. This research therefore distribution resulting from a blast and is usually
investigates the performance of the Kuz-Ram, KCO dependent on the geometry of blast.
and Modified Kuz-Ram models for fragmentation
14B   1  B   W   BCL  CCL  (3)
S 0.1
prediction using data from two Ghanaian mines.   L
n   2.2     1   abs    0.1
 d   2   B   L   H
 
The performance of these models were assessed
using the root mean square error (RMSE) and where B is burden, m; S is spacing, m; d is the hole
correlation and regression analysis. Fragmentation diameter, mm; W is the standard deviation of
measurement was done using the image analysis drilling precision, m; L is the charge length, m; BCL
technique. is the bottom charge length, m; CCL is the column
charge length, m; and H is the bench height, m.
2 Fragmentation Modelling
The rock factor in Eqn. 1 is estimated as:
A number of models have been established
to predict size distributions from specific A = 0.06(RMD + RDI + HF) (4)
blast designs. However in this research, the where RMD is the rock mass description; RDI is the
Kuz-Ram, KCO and Modified Kuz-Ram density influence; and HF is the hardness factor.
models have been studied.
2.2 The KCO Model
2.1 The Kuz-Ram Model
This model is a modified form of the Kuz-Ram
The model comprises three key equations: model and comprises three equations. The Rosin-
Kuznetsov’s equation, Rosin-Rammler equation and Rammler function in the Kuz-Ram model is
Uniformity index equation. replaced by the Swebrec function while the
uniformity index is replaced by the curve undulation
factor.

MP 2
Kuznetsov equation replaced by the blastability index. It is as shown in
Eqn. 8.
This equation as shown in Eqn. 5, is used in
estimating the mean fragmentation of the blast. 0.8 19
V  1 S  30
X m  0.073BI 0  Q e 6  ANFO  (8)
19  Qe   115 
 1  16  115  30
X 50  A. 0.8 .Q .  (5)
q   S ANFO  where BI is the blastability index; Vo is the volume
of material blasted per hole, m3, Qe is the quantity of
explosives per hole, kg.
where q is the powder factor, kg/m3; X50 is the mean
fragment size, cm; and SANFO is the relative weight Modified uniformity index
strength of the explosive to ANFO.
Cunningham’s uniformity index is modified as
Swebrec function shown in Eqn. 9 and used to characterise the
fragmentation distribution.
This function (Eqn. 6) replaces the Rosin-Rammler
equation in the Kuz-Ram model.
n'  1.88 * n * BI 0.12 (9)
1 where n’ is the modified uniformity index; n and BI
P(x)  (6)
  b
 are as defined in Eqns. 3 and 8 respectively.
  In  X max  

   X   Blastability index
1   
  In  X max  
  The blastability index (BI) characterises the rock
   X 50    formation being blasted. It is presented in Eqn. 10.
  
BI  0.5(RMD  JPS  JPA  RDI  HF) (10)
where P(x) is the percentage of material passing
sieve size X (%); b is the curve undulation where JPS is the joint plane spacing rating; and JPA
parameter; and Xmax is the maximum in-situ block is the joint plane angle/orientation rating. RMD,
size, cm (This value can be either the spacing or the RDI and HF are as defined in Eqn. 4.
burden).
3 Data Collection and Analysis
Curve undulation parameter
3.1 Study Areas
This parameter characterises the fragmentation
distribution. It depends on Cunningham’s Data for the study were obtained from two large
uniformity index (Eqn. 3) and the mean fragment scale gold mines in Ghana: Mine A and Mine B.
size. Explosives for blasting at both Mines are supplied
by Maxam Ghana Limited. Mine A is located in
Tarkwa in the Western Region of Ghana about 85
  X 
b  2.In2.In max .n (7) km North-West of Takoradi and 320 km South-West
  X50  of Accra, Ghana’s capital (Anon., 2011). Mining is
by conventional open pit method. It is within the
tropical rain forest of Ghana with temperature
The terms of Eqn. 7 are as defined for Eqns. 2 and
ranging from 24 °C to 36 °C, humidity from 60% to
6.
above 90% and an average annual rainfall of 1 700
mm (Al-Hassan, 2007). Mine A is within the
2.3 The Modified Kuz-Ram Model Tarkwaian group of rocks with mineralisation
confined to a 50 m thick section of the Banket
This is also a modified form of the Kuz-Ram model conglomerate unit (Griffins et al., 2002).
with some modification to the Kuznetsov equation
and uniformity index. The Rosin-Rammler function Mine B is located in the Western Region of Ghana,
is maintained as in the original Kuz-Ram model. about 280 km west of Accra, and 20 km from
Essiama. The mine is also within the tropical rain
Modified Kuz-Ram equation forest with day temperatures up to 30 °C and night
temperatures between 6 and 10 °C. The average
This is a modified form of the Kuznetsov equation. annual humidity is 80% while the annual rainfall
A factor of 0.073 is introduced and the rock factor is averages 2 023 mm. The mineralisation falls within
the Birimian supergroup which has minor granitic

MP 3
intrusions bounded by large granitoid bodies to the Uniaxial compressive strength (UCS) and bulk
east and west (Johnson et al., 2012). Mining is by density tests were conducted on rock samples from
conventional open pit method. Mine A. Twelve cubic samples of dimensions 50
mm × 50 mm × 50 mm were used for the UCS test
3.2 Data Collection while 24 irregular samples were used for the bulk
density test. Geotechnical cell mapping was also
The data obtained for the research include conducted at Mine A to determine joint information
geometric, explosive and rock parameters of both and general rock description.
mines. The geometric blast parameters are presented
on Table 1 while the explosive parameters are Rock characteristics data for Mine B were obtained
shown on Table 2. The bulk explosive used at both from the Geotchnical department of the mine.
Mines is Riomex 8 000 (20% Ammonium Nitrate Summary of the rock data for Mines A and B are
Porous Prills, ANPP and 80% emulsion). The presented in Table 3. Three blasts were studied from
average density is 1.2 g/cm3 (1 200 kg/m3) and the Mine A while 4 blasts were considered in Mine B.
average VOD is 4 900 m/s. The relative weight
strength of Riomex 8 000 is 83%.

Table 1 Geometric Blast Parameters of Mines A and B


Mine A Mine B
Parameter
Blast 1 Blast 2 Blast 3 Blast A Blast B Blast C Blast D
Spacing (m) 5.5 5.5 5.5 3.5 3.5 3.2 3.2
Burden (m) 5 5 5 3 3 2.9 2.9
Bench height (m) 6 6 6 6 6 6 6
Hole diameter (mm) 165 165 165 115 115 115 115
Average Hole depth (m) 7 6.64 6.85 7.2 7 7.2 7
Average Subdrill length (m) 1 0.64 0.85 1.2 1 1.2 1
Average Final Stemming height (m) 3 3 3 3 3 3 3
Table 2 Explosive Parameters of Mines A and B
Mine A Mine B
Parameter
Blast 1 Blast 2 Blast 3 Blast A Blast B Blast C Blast D
Average Quantity per hole
102.65 93.41 98.8 51.03 49.77 64.03 64.59
(kg/hole)
Volume blasted per hole
165 165 165 63 63 55.68 55.68
(bcm/hole)
Powder factor (kg/m3) 0.62 0.57 0.60 0.81 0.79 1.15 1.16
Loading density (kg/m) 25.66 25.66 25.66 12.47 12.47 12.47 12.47
Average Charge length (m) 4.0 3.64 3.85 4.2 4 4.2 4

Table 3 Rock Parameters of Mines A and B


Parameter Mine A Mine B
Uniaxial Compressive Strength (MPa) 43.56 41.8
Bulk density (tonnes/m3) 2.39 2.8
Joint spacing (m) 1.56 2.5
Moderately strong and blocky Moderately strong and blocky
General rock description
with slight weathering with slight weathering

Images of blasted muckpiles were also sampled for considered as boulders. At Mine B, fines are defined
fragmentation analysis. They were obtained using a as fragments with size less than 10 cm while
digital camera. Location of the images within the boulders are considered as particles with size greater
muckpiles was carefully chosen using visual than or equal to 80 cm. However, for the easy
analysis technique to ensure that they were comparison with the output from Split Desktop,
representative of the muckpile. Scaling objects (two fines are taken to be 15.24 cm at Mine A and 10.16
metallic poles) of length 1.5 m were used in each cm at Mine B.
image as reference sized objects.
3.3 Data Analysis
At Mine A, any fragments with size less than or
equal to 15 cm are considered as fines while those Fragmentation prediction was done for all the blasts
with size greater than or equal to 100 cm are considered using the Kuz-Ram, Modified Kuz-Ram

MP 4
and the KCO models. An MS Excel model was built for each blast. The procedure for the analysis is
for the prediction. summarised in Fig. 1.

The images of the muckpiles obtained were analysed After fragmentation prediction and analysis, the
using Split Desktop 3.1 software and the results used Root Mean Square Error (RMSE) (using MS Excel
as the basis for the model comparison. Five images 2013) and correlation and regression analysis (using
were analysed and combined to represent the results Minitab 16 software) were used to determine model
performance.

a) Image Acquisition b) Scale Setting c) Auto Delineation (d) Manual Editing/Delineation


Fig. 1 Image Analysis Procedure in Split Desktop

4 Results and Discussions

4.1 Fragmentation Analysis


The results of the fragmentation analysis for Mine A
obtained from Split Desktop 3.1 software are
summarised in Figs. 2 and 3 while the fragmentation
results for Mine B are summarised in Figs. 4 and 5.

From Fig. 2, Blasts 1, 2 and 3 had 13.32%, 13.24%


and 13.10% of fines (< 15.24 cm) produced
respectively. Measured mean fragment sizes (F50, Fig. 3 Summary of Fragmentation Analysis for
50% passing) of 38.17 cm, 51.49 cm and 46.96 cm Mine A
were obtained for Blasts 1, 2 and 3 respectively (Fig.
3). The top size (99.95% passing) of all the blasts About 27.75%, 24.02%, 29.36% and 37.90% of
was less than the crusher gape of 100 cm (Fig. 3). fines (< 10.16 cm) were produced from Blasts A, B,
This implies that insignificant (< 0.05%) quantity of C and D respectively (Fig. 4). The measured mean
boulders could be produced from all the blasts. fragment sizes (50% passing) for Blasts A, B, C and
Hence, the fragmentation results were satisfactory D respectively are 20.61 cm, 14.83 cm, 20.16 cm
with insignificant or no boulders. and 26.88 cm (Fig. 5). Only Blast B had a top size
(99.95% passing) less than the crusher gape (80 cm)
implying insignificant boulder production. Also,
over 10% of the muckpile were boulders as the F90
(90% passing) was above the crusher gape. Less
than 10% of boulders were produced from Blasts A
and C as these had an F90 below the crusher gape
(Fig. 5).

Overall, the results of the four blasts studied at Mine


B reveal that there were high amounts of fines
(averagely 29.76%) with less than 10% of boulders
except Blast D.
Fig. 2 Fragmentation Results for Mine A

MP 5
top size (99.95% passing) was less than the crusher
gape (100 cm). This indicates that the quantity of
boulders predicted by each of the models exceeded
the actual amount produced (if any at all) hence
there was over prediction of the amount of boulders
produced in Blast 1.

The predicted mean fragment sizes for Blast 1 as


shown on Fig. 9 are less than the measured mean
fragment size (F50) for Blast 1 which is 38.17 cm
(Fig. 3). The prediction errors for the mean fragment
Fig. 4 Fragmentation Results for Mine B size are 25.62%, 32.93% and 25.33% for the Kuz-
Ram, KCO and Modified Kuz-Ram models
respectively.

Fig. 5 Summary of Fragmentation Analysis for


Mine B
Fig. 7 Fragmentation Prediction Summary for
4.2 Fragmentation Prediction Blast 2 of Mine A

The fines predictions by the three models as


The fragmentation prediction results using the Kuz-
presented on Fig. 7 exceeded the actual quantity of
Ram, Modified Kuz-Ram and KCO models for
fines produced from Blast 2 (13.24%). This results
Mine A are summarised in Figs. 6 to 9. The results
in fines prediction errors of 52.28%, 56.97% and
for Mine B are presented in Figs. 10 to 14.
49.47% for the Kuz-Ram, KCO and Modified Kuz-
Ram models respectively. Similar to Blast 1, there is
over prediction of the quantity of fines by all the
models. Also, there was over prediction of the
boulder quantities produced in Blast 2 as the
predicted boulder amounts (Fig. 7) exceeded the
actual quantity of boulders produced (Fig. 2).

The predicted mean fragment sizes for Blast 2 as


presented on Fig. 9 exceeded the actual mean
fragment size for this blast (51.49 cm). There was
therefore under prediction of the mean fragment size
by the Kuz-Ram, KCO and Modified Kuz-Ram
Fig. 6 Fragmentation Prediction Summary for models with errors of 42.42%, 48.07% and 42.20%
Blast 1 of Mine A respectively.
The actual quantity of fines produced from Blast 1
was 13.32% (Fig. 2). The Kuz-Ram model predicted
25.43% as fines (< 15.24 cm), KCO predicted
28.59% fines while the Modified Kuz-Ram model
predicted 23.79% as fines (Fig. 6) to be produced
from Blast 1. Hence, the models actual amount of
fines was exceeded by the models’ predictions with
errors of 47.62%, 53.41% and 44.01% respectively.

The quantity of boulders predicted for Blast 1 are


shown on Fig. 6. The actual quantity of boulders
produced from Blast 1 was less than 0.05% as the Fig. 8 Fragmentation Prediction Summary for
Blast 3 of Mine A

MP 6
Similar to Blasts 1 and 2, all the models over The quantity of fines predicted to be produced from
predicted the quantity of fines produced in Blast 3 as Blast (Fig. 10) were less than the measured amount
the predicted quantity of fines (shown on Fig. 8) of fines from this blast (24.04%) (Fig. 4). This
exceeded the measured quantity of fines from this implies that all the models under predicted the
blast (13.15%). The fines prediction errors for Kuz- amount of fines produced with errors of 16.56%,
Ram, KCO and Modified Kuz-Ram models are 5.16% and 18.93% respectively for the Kuz-Ram,
48.47%, 54.09% and 44.96% respectively. KCO and Modified Kuz-Ram models.

The boulder amounts predicted by all the models The boulder (> 80 cm) predictions for all the models
(Fig. 8) were greater than the actual quantity of were less than 3% (Fig. 10). This blast had a little
boulders produced since the top size (99.95% over 10% boulders since the F90 (81.31 cm) (Fig. 5)
passing) of Blast 3 was 94.61 cm (less than the was higher than the crusher gape (80 cm). This
crusher gape) as shown on Fig. 3. All the models implies that there was an under prediction of the
thus over predicted the quantity of boulders boulder amounts by all the models considered.
produced from this blast.
The mean fragment sizes predicted by the models
The predicted mean fragment sizes for Blast 3 as are summarised on Fig. 14. The actual mean
presented on Fig. 8 are all less than the measured fragmentation (F50) of the blast was 26.88 cm
mean fragment size of 46.96 cm (Fig. 3). Hence implying an under prediction of the mean fragment
there was under prediction of the mean fragment size with errors of 17.56%, 25.63% and 20.35% for
sizes with errors of 38.88%, 44.88% and 38.65% for the Kuz-Ram, KCO and Modified Kuz-Ram models
the Kuz-Ram, KCO and Modified Kuz-Ram models respectively.
respectively.

Fig. 9 Mean Fragment Sizes for Mine A Fig. 11 Fragmentation Prediction Summary for
Blast B of Mine B
Overall, the predictions from Blasts 1, 2 and 3 of
Mine A show a general trend of over estimation of Under prediction of fines as observed in Blast A is
both fines (< 15.24 cm) and boulders (> 100 cm) by also observed in Blast B as all the models predicted
all the models considered. This finding contradicts less than 33% fines (Fig. 11) while the measured
what was reported by Esen (2013) and Cho and quantity of fines is 37.90% (Fig. 5). The prediction
Kaneko (2004) that the Rosin-Rammler function errors for Kuz-Ram, KCO and Modified Kuz-Ram
(used in the Kuz-Ram model) underestimates the models are 23.59%, 14.22% and 23.67%
amount of fines produced. The models also respectively.
performed poorly in predicting the fines as the fines
prediction errors were generally high (> 40%). Insignificant amount of boulders (< 1%) were
predicted to be produced from this blast (Fig. 11).
From the fragmentation analysis, less than 0.05% of
boulders could be produced from this blast as the top
size was below the crusher gape of 80 cm. This
implies that there is over prediction of the boulder
quantities produced from this blast.

The mean fragment sizes predicted by all the models


(Fig. 15) exceeded the measured mean fragment size
of this blast (14.83 cm). All the models therefore,
over predicted the mean fragment sizes with errors
of 51.25%, 36.41% and 46.06% for the Kuz-Ram,
KCO and Modified Kuz-Ram models respectively.
Fig. 10 Fragmentation Prediction Summary for
Blast A of Mine B

MP 7
29.36% (Fig. 4). While the Kuz-Ram and Modified
Kuz-Ram models under predicted the amount of
fines by 0.75% and 0.72%, the KCO model over
predicted the quantity of fines by 10.38%.

As shown on Fig. 13, the boulder predictions (< 1%)


were all less than the actual quantity of boulders
produced from this blast (< 10%) as the top size of
this blast (101.17 cm) is greater than the crusher
gape while the F90 is less than the crusher gape.
There was therefore under prediction of the boulder
Fig. 12 Fragmentation Prediction Summary for amounts by all the models considered.
Blast C of Mine B
The mean fragment size for Blast D was under
The Kuz-Ram, KCO and Modified Kuz-Ram predicted by all the models as shown on Fig. 15. The
models predicted 19.48%, 22.17% and 18.89% of prediction errors were 18.08%, 26.08% and 20.86%
fines respectively to be produced from Blast C of for the Kuz-Ram, KCO and Modified Kuz-Ram
Mine B (Fig. 12). The actual amount of fines models respectively.
produced however, was 27.75% (Fig. 4). There is
therefore an under prediction of the actual amount of
fines from all the models with errors by the Kuz-
Ram, KCO and Modified Kuz-Ram models of
29.80%, 20.11% and 31.93% respectively.

The boulder predictions by all the models were less


than 1% (Fig. 12). Less than 10% of boulders were
produced from this blast as the F90 was less than the
crusher gape but the top size was greater crusher
gape. There was therefore an under prediction of the
boulder quantities produced from this blast.
Fig. 14 Mean Fragment Sizes for Mine B
The mean fragment size for this blast was under
predicted by all the models as shown on Fig. 14. The Overall, the blasts from Mine B agreed with the
prediction errors were 18.78%, 26.73% and 21.54% general trend of fines underestimation by the Kuz-
for the Kuz-Ram, KCO and Modified Kuz-Ram Ram model (Esen, 2013; Cho and Kaneko, 2004) as
models respectively. this trend was observed for all the blasts at Mine B.
There was also a general trend of underestimation of
the quantity of boulders produced from the blasts at
Mine B except Blast B for which there was slight
over prediction of the quantity of boulders.

The models performed poorly in predicting the fines


as errors were high (up to 32%). The models
however performed better in predicting the quantity
of boulders with minimal errors.

4.3 Model Performance


Fig. 13 Fragmentation Prediction Summary for A summary of the results of the RMSE
Blast D of Mine B computations using the predictions from the models
and the fragmentation measurement from image
The Kuz-Ram, KCO and Modified Kuz-Ram analysis for all the blasts at Mine A and Mine B are
models predicted 29.14%, 32.76% and 29.15% (Fig. shown in Tables 5 and 7 respectively. Tables 6 and
13) of fines respectively to be produced from Blast 8 summarise the correlation and regression analysis
D. The actual amount produced however, was results.

MP 8
Table 5 Estimated RMSE of Models for Mine A
Model RMSE (%)
Blast Best Predictor
Kuz-Ram KCO Modified Kuz-Ram
Blast 1 7.74 9.8 7.34 Modified Kuz-Ram
Blast 2 10.82 12.84 10.67 Modified Kuz-Ram
Blast 3 9.51 11.55 9.47 Modified Kuz-Ram

Table 6 Summary of Correlation and Regression Analysis Results for Mine A


Model R (%)
Blast Best Predictor
Kuz-Ram KCO Modified Kuz-Ram
Blast 1 98.5 97.7 98.6 Modified Kuz-Ram
Blast 2 97.0 96.2 97.1 Modified Kuz-Ram
Blast 3 97.9 97.0 97.9 Kuz-Ram/Modified Kuz-Ram

The results from the RMSE computation have the highest correlation coefficient (R) for all the blasts
same trend for all the blasts at Mine A. The Modified with the KCO model having the least R values for
Kuz-Ram model had the least RMSE followed by all the blasts at Mine A. The Modified Kuz-Ram
the Kuz-Ram model while the KCO recorded the model is therefore the model with the best
highest RMSE (Tables 11). Same trend is observed fragmentation prediction performance for Mine A.
using the correlation and regression analysis results The other models (Kuz-Ram and KCO) however,
(Table 6) as the Modified Kuz-Ram model had the show strong correlation with R values above 95%.

Table 7 Estimated RMSE of Models for Mine B


RMSE (%)
Blast Best Predictor
Kuz-Ram KCO Modified Kuz-Ram
Blast A 6.41 7.34 7.58 Kuz-Ram
Blast B 11.10 8.96 10.70 KCO
Blast C 6.00 7.28 7.33 Kuz-Ram
Blast D 6.83 7.66 7.94 Kuz-Ram

Table 8 Summary of Correlation and Regression Analysis Results for Mine B


Model R (%)
Blast Best Predictor
Kuz-Ram KCO Modified Kuz-Ram
Blast A 99.1 98.7 98.7 Kuz-Ram
Blast B 98.4 99.1 98.6 KCO
Blast C 99.3 98.9 98.9 Kuz-Ram
Blast D 99.1 98.7 98.7 Kuz-Ram

From Table 13, the Kuz-Ram model had the least confirms that, the performance of fragmentation
RMSE for all the blasts at Mine B except Blast B for prediction models are site specific as observed by
which the KCO model had the least RMSE. Also, Faramarzi et al., (2013).
the Kuz-Ram model had the highest R values for all
the blasts at Mine B except Blast B for which the 5 Conclusions and Recommendations
KCO had the highest R. The Kuz-Ram model
therefore, had the best prediction performance at 5.1 Conclusions
Mine B.
The average actual quantity of fines produced from
As observed at Mine A, all the models had high R the three blasts studied at Mine A was 13.24% with
values (above 95%). This indicates that the insignificant amount of boulders while at Mine B,
predictions from all the models correlate well with the blasts produced up to 37.90% of fines with
the actual results from the blasts studied and hence insignificant amount of boulders.
may also be used for fragmentation prediction and
blast optimisation at the mines. The study shows a general trend of over estimation
of fines (< 15.24 cm) and boulders (> 100 cm) for
Also, the formations studied showed different trends the blasts at Mine A and underestimation of fines (<
in the models’ prediction performance. This

MP 9
10.16 cm) and boulders (> 80 cm) at Mine B for the Choudhary, B. S. and Gupta, A. (2012), “A
models considered. Also, the mean fragment sizes at Comparative Study on Fragmentation
both mines were underestimated by all the models. Measurement Techniques”, Mining Engineers’
Journal, Mining Engineers’ Association of
All the models had weak performance in predicting India, Vol. 14, No. 3, pp. 16 – 24.
the quantity of fines for both mines. This implies
Cunningham, C. V. B. (2005), “The Kuz-Ram
that the modifications made on the Kuz-Ram model
Fragmentation Model – 20 Years on”,
to arrive at the KCO and Modified Kuz-Ram models
Proceedings of the Brighton Conference 2005:
have not led to a universal improvement in the
European Federation of Explosives Engineers,
prediction of fines. All the models performed better
Halmberg, R. et al (ed.), Brighton, Sussex,
in predicting the quantity of boulders at both mines
England, pp. 201 – 210.
as the prediction errors were generally less than
10%. Djordjevic, N. (1999), “Two-Component Model of
Blast Fragmentation”, Proceedings of Fragblast
The correlation coefficients (R) of all the models conference, South African Institute of Mining
were generally high (above 95%) indicating strong and Metalurgy, Johannesburg, pp. 213 – 219.
correlation and good model prediction performance.
Engin, I. C. (2010), A Practical Method of Bench
However, Modified Kuz-Ram model performed best
Blasting for Desired Fragmentation based on
at Mine A while Kuz-Ram model performed best at
Digital Image Processing Technique and the
Mine B in predicting the fragmentation distribution.
Kuz-Ram Model, Afyon Kocatape University,
The formations studied showed different trends in
the models’ prediction performance indicating that Iscehisar Vocational School, Iscehisar,
models’ performance is site specific. Afyonkarahisar.
Faramarzi, F., Mansouri, H. and Ebrahimi, F. M. A.
5.2 Recommendations (2013), “A Rock Engineering Systems Based
Model to Predict Rock Fragmentation by
The Modified Kuz-Ram model is recommended for Blasting”, International Journal of Rock
fragmentation prediction and blast optimisation at Mechanics and Mining Sciences, Vol. 60, pp. 82
Mine A while the Kuz-Ram model is recommended – 94.
at Mine B. Since the results of the predictions from
Gheibie, S., Aghababaei, H., Hoseinie, S. H. and
all the models had strong correlation, they may be
Pourrahimian, Y. (2009), “Modified Kuz - Ram
considered when designing blasts for greenfields
Fragmentation Model and its use at the Sungun
with sufficient knowledge of the rock properties.
Copper Mine”, International Journal of Rock
This proves superior to the trial and error methods
Mechanics & Mining Sciences, Vol. 46, pp. 967
commonly used in industry.
– 973.
References Griffins, R. F., Barning, K., Agezo, F. L. and Akosa,
F. K. (2002), Gold Deposit of Ghana, Minerals
Afum, B. O. (2014), “Principles of Mining”, Commission, Accra, Ghana, pp. 118 – 132.
Unpublished Short Course Lecture Notes,
University of Mines and Technology, Tarkwa, Johnson, C. E. (2014), “Fragmentation Analysis in
74 pp. the Dynamic Stress Wave Collision Regions in
Bench Blasting”, Unpublished PhD Thesis
Al-Hassan, S. (2007), “A Comparative Study of Report, University of Kentucky, U.S.A, 158 pp.
Rainfall in Tarkwa Mining District of Ghana”,
Ghana Mining Journal, Vol. 9, pp. 25 – 32. Johnson, N., De Klerk, Q., Yeo, W. and Roux, A.
(2012), “Technical Report and Mineral Resource
Anon. (2011), “Mineral Resource and Ore Reserve: and Reserve Update for the Nzema Gold Mine,
Report Pure Gold”, AngloGold Ashanti, Ghana, West Africa”, Report Prepared for
www.anglogoldashanti.com, 184 pp. Accessed: Endeavour Mining Corporation, Cayman
August 4, 2015. Islands, 205 pp.
Anon. (2015), “AngloGold Ashanti Iduapriem Mine Kanchibotla, S. S., Valery, W. and Morrell, S.
– Mining Department”, Unpublished Induction (2001), “Modelling Fines in Blast Fragmentation
PowerPoint Slides, Tarkwa, Ghana, 49 pp. and Its Impact on Crushing and Grinding”,
Cho, S. H. and Kaneko, K. (2004), “Rock Julius Kruttschnitt Mineral Research Centre
Fragmentation Control in Blasting”, 2004 Internal Report, University of Queensland,
Materials Transactions: The Mining and Australia, 20 pp.
Materials Processing Institute of Japan, Vol. 45, Kazem, O. and Bahareh, A. (2009), Prediction of
No. 5, pp. 1722 – 1730. Rock Fragmentation in Open Pit Mines, using
Neural Network Analysis, 15 pp.

MP 10
Ouchterlony, F. (2005), “The Swebrec© Function: Mining Systems, Operations Research in Mining, Explosives and
Rock Fragmentation, Blast Optimisation, Environmental Impacts
Linking Fragmentation by Blasting and
of Blasting and Small Scale Mining.
Crushing”, Mining Technology (Trans. Instn.
Min. Metall. A), Vol. 114, pp. A29 – A44. Victor A. Temeng is an Associate Professor
in Mining Engineering and the Dean of the
Siddiqui, F. I., Shah, S. M. A. and Behan, M. Y. School of Postgraduate Studies at the
(2009), “Measurement of Size Distribution of University of Mines and Technology
Blasted Rock Using Digital Image Processing”, (UMaT), Tarkwa. He obtained his BSc
JKAU: Eng. Sci., Vol. 20, No. 2, pp. 81 – 93. (Hons.) and Postgraduate Diploma in Mining
Engineering degrees from UMaT in 1985
Vamshidhar, K. and Venkatesh, H. S. (2010), and 1986 respectively. He obtained his MSc
degree from the University of Zambia in 1992 and his PhD degree
“Review of Models for Prediction of Rock from the Michigan Technological University, USA in 1997. He
Fragmentation due to Blasting”, Journal of the joined UMaT as a Lecturer in 1990. He is a member of the Society
Explosives Safety And Technology Society of Mining, Metallurgy and Exploration (MSME) and a member
(Visfotak) India: Dealing With Safety And of the Ghana Institution of Engineers (GhIE). His areas of
specialisation include Operations Research, Materials Handling
Technological Aspects Of The Explosives
and Computer Applications.
Industry, Vol. 5, pp. 23 – 30.
Vasileios, D. (2008), “Fragmentation Analysis of Bright Oppong Afum is a Lecturer in the
Optimised Blasting Rounds in the Aitik Mine – Mining Engineering Dept. of the University of
Effect of Specific Charge”, Unpublished Mines and Technology (UMaT). He holds
degrees in BSc Mining Engineering from the
Masters Thesis Report, Luleå University of Kwame Nkrumah University of Science and
Technology, Sweden, 108 pp. Technology (KNUST), and MSc
Environmental Monitoring and Analysis from the Aberystwyth
University, UK. He has worked in an explosives and underground
Authors mining companies before joining UMaT. He is a member of the
Society for Mining, Metallurgy and Exploration (MSME), the
Kansake Bruno Ayaga is currently an Australasian Institute of Mining and Metallurgy (MAusIMM)
MPhil. Candidate and and the Ghana Institution of Engineers (GMGhIE). He is
Demonstrator/Teaching Assistant at the currently a PhD researcher at UMaT.
University of Mines and Technology
(UMaT), Tarkwa, Ghana. He holds a
BSc (Honours) from UMaT. His
research interests are in Simulation of

MP 11

View publication stats

Você também pode gostar