Professional Documents
Culture Documents
By
+
++
This paper is to be published by SME in the proceedings of Control 2000 Symposium to held in
conjunction with the 2000 SME Annual Meeting and Exhibit, February 28-March 1, Salt Lake City, Utah.
ABSTRACT
Optimizing blasting, crushing and grinding operations is filled with challenges.
One of the more difficult tasks is accurately sampling and determining the size
distributions of blasted and crushed materials at a reasonable cost. The task is difficult
because of the large size of fragments and the tonnage involved. However, the size
distribution measurements are necessary for models that predict the performance of mine
through mill operations. This paper is concerned with the use of video sampling for this
task at National Steel Pellet Company's operation in Keewatin, Minnesota. Data
gathering, data analysis, model building and mine-to- mill simulation are all described.
INTRODUCTION
Mining companies around the world are seeking ways to optimize performance.
In recent days a great deal of attention is being paid to optimizing the mine/mill interface
(Morrell, 1998). A principal challenge in carrying out such an optimization is to measure
the performance of blasting operations, crushing operations, and primary grinding
operations reliably and inexpensively. Measuring fragment size distributions at each
stage of the size reduction process is critical in order to establish a baseline for predictive
simulators to use in calibration, and for evaluation of process improvements.
Unfortunately, fragment sizes in muck piles, trucks, crusher dump stations, and on
primary mill feed conveyors are large and highly variable making conventional sampling
and screening at least expensive, and in some instances impossible.
National Steel Pellet Company (NSPC) is continuously seeking to optimize its
mine/mill performance through ore blending at the mine and adaptive fine-tuning through
control. The company operations are located in Northern Minnesota in the town of
Keewatin. Annually it processes about 18 M tonnes of taconite ore to produce
approximately 5.35 M tonnes of iron ore pellets. The ore characteristics for different
locations in the NSPC Pit are quite variable. The flowsheet for mining and grinding
portions of the operation are shown in Figure 1. Blasting is currently accomplished using
ammonium nitrate emulsion-based blasting agents. Ore is loaded into trucks (eight 240tonne, four 205-tonne) and hauled about one mile to two 1.524 m x 2.59 m (60" x 102")
primary gyratory crushers driven by one 600 kW (800 hp) motor and one 675 kW (900
hp) motor. The crushed product is conveyed to a 220,000 tonne coarse ore storage barn.
In turn, the ore in storage is conveyed to ten 8.23 m x 5.49 m (27' x 18') SAG mills which
are each driven by two 2625 kW (3500 hp) motors.
++
Loading
Blasting
Drilling
Hauling
Crushing
Storage
Primary Milling
Figure 1. NSPC mine-to-mill operations.
NSPC working in conjunction with J.A. Herbst & Associates has recently been
evaluating the use of video sampling to measure blasting, crushing, and grinding behavior
of different ores. The sampled images are analyzed with transformed video image
software. The resulting fragment size distributions are used to calibrate a mine-to-mill
flowsheet simulator. This paper describes the video sampling process and the simulator
calibration. Finally, some illustrations of the potential usefulness of the data and
simulator are presented.
VIDEO SAMPLING
Video sampling was accomplished using a JVC Mini Digital Video Camera (GRDVM5) with a 100X zoom. Truck contents were sampled by collecting video images of
material in the bed of four separate trucks over the entire time each truck was dumping
into a primary crusher. A reference size for truck images was established based on the
known width of truck tires. Products from the two crushers were sampled by placing the
camera over high-speed conveyors carrying the crushed material to a tripper system for
distribution in the ore storage barn. Finally, SAG feed was measured by placing the
camera over three of the primary mill feed conveyors. Reference sizes on conveyor belts
were established using wooden dowel pieces cut to a length of 25 mm each. A shutter
speed of 1/500th of a second was used for all sampling. Natural light was used for
outdoor taping of the trucks at the crusher while auxiliary artificial light was provided for
indoor taping.
Raw images were transferred from tape to an IBM 385XD laptop through a Video
Port Pro frame grabber. The raw images were then analyzed using the OPSA Software
developed at University of Utah (Miller, 1999). This software makes a series of
enhancements and transformations on each image. The first of these enhancements are
4
shown for the case of one image obtained at the beginning of a dump of truck 4292. Here
the raw image captured from the videotape is enhanced by brightening. See Figure 2a and
2b. Edge finding is then used to prepare the image for chord length distribution
measurements shown in Figure 2c. The resulting surface chord length distribution is
shown plotted in Figure 3. The OPSA software then makes the stereological
transformation from the linear chord distribution to the volumetric distribution of
particles in the exposed or surface layer of the truck as shown in Figure 3. The
transformation from the volumetric distribution of the exposed layer to the desired
volumetric distribution of particles in the bulk of the truck is also shown in Figure 3.
Original Image
Brightened Image
Separated Image
100
100
Truck 4292 at beginning of dump
Measured chord length
distribution for surface
80
60
60
Transformed volume
distribution for bulk
40
40
20
20
0
1
10
100
Number % Finer
Weight % Finer
80
0
1000
Size, mm
Figure 3. Transformation of chord length distribution to volumetric size distribution for the bulk.
Since each image contains only a finite number of fragments, the statistics of
counting are important. For this reason, five separate raw images from the beginning of
the dump were analyzed and the resulting size distributions averaged. This procedure
was repeated for five images in the middle of the dump and five more at the end of the
dump. The overall average of the beginning, middle and end images is shown in Figure
4. The differences between the average size distribution for the bulk at the beginning,
middle and end are relatively small. In contrast, the overall size distributions of the four
trucks from different loading locations varied strongly as shown in Figure 5.
100
Truck 4292
80
Percent Passing
Beginning
Middle
60
End
Avg
40
20
0
1
10
100
1000
Size, mm
Figure 4. Size distribution of single truck from average of several images during dump.
100
Truck 4292
Percent Passing
80
Truck 4296
Truck 4298
60
Truck 4293
40
20
0
1
10
100
1000
Size, mm
Figure 5. Variation in size distribution from truck to truck.
The image analysis methodology for the conveyor video sampling was identical
to that for the trucks described above. However, the overall analysis of conveyor size
distributions did differ, because there are fewer fragments per image as is seen by
comparing Figure 6 and Figure 2. The standard deviation of any counting procedure is
inversely proportional to the square root of the number of things counted. Figure 7 shows
a plot of estimated standard deviation versus 1 N for different numbers of images (N)
on the belt. Due to the unfavorable statistics of counting conveyor images, the overall
conveyor size distributions were determined by averaging 80 images rather than the five
used for trucks.
100
Crusher
Percent Passing
80
60
40
20
0
1
10
100
1000
Size, mm
40
N=80
N=20
HTG
N=5
SX
30
Rock
20
Fines
10
0
0.0
0.1
0.2
0.3
N
0.4
0.5
0.6
-0.5
Figure 7. Effect of number of images on standard deviation of mean size distribution measurements.
The results of the size distributions determined in this fashion from Crusher 1
(which had a worn mantle) and Crusher 2 (which had a new mantle) are shown in Figure
8. Even though the two crushers were operated with "identical" open side settings of 200
mm (8") the products are seen to be quite different.
100
Crusher #1
Percent Passing
80
Crusher #2
60
New mantle
40
Worn mantle
20
0
10
100
1000
Size, mm
Figure 8. Evaluation of performance differences between crushers.
The discharge from either crusher is distributed into 10 piles by two trippers in the
ore storage barn. Each mill receives feed from its own pile with one or two pan feeders
emptying onto the conveyor belt. Figure 9 shows that after averaging 80 images there are
significant differences in the average size distribution to each mill. Figure 10 shows
moving average values calculated from images on one line. The data indicates that each
mill experiences significant variations in the feed size distribution over time. These
observations are particularly important, since it is known that some media pieces in the
feed (+100 mm) are required to achieve to good SAG throughput, while a large amount
of hard-to-grind (50x 100 mm) in the feed limits mill capacity.
100
Line 2
Line 4
Percent Passing
80
Line 7
60
40
20
0
1
10
100
1000
Size, mm
Figure 9. Comparison of size distributions for 3 SAG feed lines.
40
50
Minus 25 mm
40
50 x 100 mm
20
30
10
Weight %
Weight %
30
20
Plus100 mm
0
0
30
60
90
Time, min
120
150
10
180
Figure 10. Time variation of rocks, hard-to-grind and fines in SAG feed to one mill.
With regard to accuracy of the size distributions (i.e. how closely they match
screen size analyses), Figure 11 shows that the screen analysis of a four ton sample of
SAG feed is very close to OPSA/T-VIS* volumetric distribution determined for the bulk.
100
Cumulative % Passing
80
60
40
Sieve Analysis (Bulk)
OPSA/T-VIS (Bulk)
20
0
1
10
100
1000
Particle Size, mm
T-VIS is the commercial name of the video imaging system containing OPSA software sold
under license by J. A. Herbst & Associates, LLC.
MODEL CALIBRATION
Important models for the simulation of the NSPC mine-to- mill interface are an
explosive breakage model, a primary crusher model, and a SAG mill model plus auxiliary
transport and storage models. The models used in this investigation were selected from
those provided in the dynamic flowsheet simulator MinOOcad (Herbst & Pate, 1998).
Most of the parameters for these models are the physical variables such as equipment
dimensions and settings that are known. The ore variables are the only ones that must be
estimated from performance data. MinOOcad provides a set of reference or default
parameters for a "typical" taconite ore. Using these parameters as starting values, model
calibration is relatively easy, involving the adjustment of a single calibration constant for
each unit operation; e.g. an explosive index, EI; a crusher index, CI; a SAG rock
competency index, SRI; SAG hard-to-grind index, SHI; and SAG particle index, SPI.
Figure 12 illustrates the calibration procedure for the explosive breakage model.
The adjusted value of EI = 9.5 kWh/MT gives good agreement between the experimental
size distribution of the fragments from truck and the calibrated explosive model and is,
therefore, deemed the best estimate for this ore. Figures 13 and 14 show similar
comparisons of experimental distributions from video sampling and the corresponding
MinOOcad model fits for the explosive breakage model and the crusher model.
100
80
Percent Passing
PF = 0.177
EI = 10.5
60
EI = 9.5
40
EI = 8.6
Measured
20
0
1
10
100
1000
Size, mm
10
100
100
Percent Passing
Percent Passing
Truck 4296
Truck 4292
80
60
40
80
60
40
20
20
0
1
10
100
1000
10
Size, mm
100
1000
Size, mm
100
100
Truck 4293
Percent Passing
80
Percent Passing
Truck 4298
Model, EI = 8.1 kWh/mt
60
40
80
60
40
20
20
0
1
10
100
1000
10
100
1000
Size, mm
Size, mm
Figure 13. Best fit explosive breakage model size distributions for 4 ore types.
100
Crusher #2
Percent Passing
80
Measured
CI=9.4 kWh/mt
60
40
20
0
1
10
100
1000
Size, mm
Figure 14. Best fit of crusher model for blended ore feed.
confirmation in which predicted and measured feed and products from the crushers are
compared for a mix of the four ore types.
100
Percent Passing
Predicted Product
80
Measured Product
Predicted Feed
60
Measured Feed
40
20
0
1
10
100
1000
Size, mm
12
Ore Feedrate
MTPH
Total Energy
KWh/MT
285
18.7
287
312
326
334
18.6
17.1
16.4
16.0
326
326
326
16.5
16.4
16.3
One of the real advantages of the simulation evaluation is that results can be
understood in fundamental terms. Blending obviously reduces variations in tonnage,
making it unnecessary to cap tonnages for soft ores which can overload downstream
processes, or to run equipment very near power limits when hard ore is processed. The
net result is that at the same crusher setting (OSS = 200 mm), the blend can be processed
at 326 mtph rather than the average of 285 mtph when processed separately. The effect
of increasing the crusher OSS may at first seem counter intuitive (more finely crushed
feed requiring more energy in the SAG mills). However, the reason becomes apparent if
one examines Figures 17 and 18. Here we see that coarser crushing provides more rocks
13
and associated media pieces (+100 mm) relative to the intermediate, hard to grind (HTG)
fractions (50 x 100 mm) and fines (-50 mm). As the ratio of hard to grind pieces to
media rocks in the mill becomes more favorable (lower), the grinding rate increases,
yielding a higher feedrate at the same filling (26.7% volume filling of ore and balls). The
simulations predict that this benefit becomes marginal as the crusher is opened beyond
225 mm probably because media rocks begin to take up too much space in the mill.
100
Percent Passing
150 mm
175 mm
200 mm
225 mm
Crusher Product
80
60
OSS
40
20
Fines
HTG Rocks
0
1
10
100
1000
Size, mm
340
Throughput, MTPH
320
310
300
290
280
Primary Mill
330
270
260
125
150
175
200
225
0
250
Figure 18. Effect of open side setting on throughput and ratio of hard-to-grind to media pieces.
Figure 19 shows that given the current estimated blasting and crushing
efficiencies, one should probably minimize the amount of blasting while keeping in mind
14
that blasted material must be small enough to be loaded and fed to the crusher. In
addition, it cannot be so coarse as to exceed the power draw of the crusher. In any case,
the reductions in total energy are quite small and therefore other factors may dominate
the decision on blasting practice.
0.7
Energy, kWh/mt
0.6
Total Energy
0.5
0.4
Blasting Energy
0.3
0.2
0.1
0.0
0.00
Crushing Energy
0.05
0.10
0.15
0.20
0.25
15
820
5200
800
5000
780
4800
760
4600
740
4400
Meas Power
Est Power
4200
720
Power, kW
5400
700
0
60
120
Time, min.
180
25
240
38
20
36
15
34
Estimated Ore Filling, %
10
32
Estimated Ball Filling, %
5
0
60
120
180
Filling, %
Estimated Angle, %
30
240
Time, min.
450
0.15
Feedrate, ltph
350
0.12
Feedrate
250
150
0.09
0.06
Grindability, [kWh/lt]-1
Estimated Grindability
50
0.03
0
60
120
180
240
Time, min.
Figure 20. Softsensor estimates of SAG mill variables with predicted model based control performance.
16
CONCLUSIONS
This paper has examined video sampling as a tool for mine-to-mill optimization.
It was found that video samples of mine trucks, crusher products and SAG feed materials
collected at National Steel Pellet Company's Keewatin operations provided valuable
insight into the workings of the mine/mill interface. Image analysis of the video samples
provided accurate size analyses for mine-to-mill performance evaluation and also
produced useful input for the calibration of blasting, crushing and SAG milling models.
These calibrated models were in turn used in a mine-to-mill simulator to help identify and
evaluate promising alternatives for increasing throughput given current ore conditions.
ACKNOWLEDGEMENTS
The authors wish to thank NSPC management for permission to publish these
findings. The assistance of Mr. Don Healy and Mr. Phillip Murr during video sampling
and the help of Dr. William T. Pate and Mr. Richard T. Herbst during image analysis are
also gratefully acknowledged.
REFERENCES
Herbst, J.A. and Pate, W.T., Dynamic Simulation of Size Reduction Operations from
Mine to Mill, Mine to Mill 1998 Conference, AusIMM, October 1998, p. 243.
Herbst, J.A. and Pate, W.T., Object Components for Comminution Systems Softsensor
Design, 9th European Symposium on Comminution, Prints Volume 2, p.741.
Lin, C.L. and Miller, J.D. Plant-site Evaluations of the OPSA System for Online Particle
Size Measurements from Moving Belts, Preprints Annual SME Meeting, Denver,
Colorado 1999.
Scott, A. and Morrell, S., 1998 Mine to Mill Conference, AusIMM, October 1998.
17