University
stringclasses 19
values | Text
stringlengths 458
20.7k
|
|---|---|
Colorado School of Mines
|
5.1.3 The Effect of Missing Matches
This section discusses the results obtained from testing the effect of missing matches on
the distance calculation algorithm. The miss rates tested were described previously in Section
4.2.3.2. Additionally, this test was performed on all the results obtained from Section 5.1.1. In
other words, the image-matching algorithm is intentionally forced to miss matches from the
results obtained in the effect of the distance between the sensors test. Therefore, this test will
include both the granite rock and the white-painted wall sceneries and will consist of various
distances between the sensors and sampling criteria. In this section, the “Normal Calculation(cid:180)
label presented in the upcoming figures represents the results obtained when no misses occurred,
i.e., the normal results presented in Section 5.1.1.
5.1.3.1 Granite Rock
Figures 5-19 and 5-20 show the effect of missing matches for the granite rock experiment
when the DBS is 8.3 cm. The results presented in Figure 5-19 include all the sampling criteria
tested for each category of the matches missed. The results presented in Figure 5-20 include all
the missing matches criteria for each sampling criterion. Figure 5-20 helps explain which
sampling criterion caused the errors to increase or shift when matches were missed.
The results show that missing matches may have a wider range of errors and may cause
the average percentage error to shift in sign or become higher. In some cases, the average
percentage error decreases when the missing matches. This decrease occurs because errors in the
original experiment already existed, and missing a match may have inadvertently adjusted for
those errors. Finally, when sampling every 8.0 s provides the lowest accuracy when missing
matches. This behavior is not surprising because the algorithm captures fewer images, leading to
fewer calculated velocities and increasing the probability of encountering errors.
90
|
Colorado School of Mines
|
Figures 5-21 and 5-22 show the effect of missing matches for the granite rock test when
the DBS is 13.3 cm. The results presented in Figure 5-21 include all the sampling criteria tested
for each category of the matches missed. The results presented in Figure 5-22 include all the
missing matches criteria for each sampling criterion.
Figure 5-21 shows that missing one match every other match resulted in a data point that
had a 90% error. This result illustrates the power of misses on the distance calculation. Figure 5-
22 shows that this outlier occurred when the sampling criterion was set to 8.0 s. This observation
is again consistent with the results demonstrated in Figure 5-20. If few images are captured and
the algorithm misses matches, errors will increase depending on the miss rate. All the other
missing matches categories with lower miss rates did not significantly affect the accuracy of the
distance calculated.
Figure 5-21 Missing matches test - the effect of missing matches for the granite rock experiment
when the distance between the sensor is 13.3 cm.
92
|
Colorado School of Mines
|
140
120
100
80
Sampling at 1 image every 2 seconds
Sampling at 1 image every 4 seconds
60
Sampling at 1 image every 6 seconds
Sampling at 1 image every 8 seconds
40
20
0
-20
Figure 5-22 Missing matches test - the effect of sampling criterion and matches missed for the
granite rock experiment when the distance between the sensor is 13.3 cm.
Figures 5-23 and 5-24 show the effect of missing matches for the granite rock experiment
when the DBS is 15.0 cm. The results presented in Figure 5-23 include all the sampling criteria
tested for each category of the matches missed. The results presented in Figure 5-24 include all
the missing matches criteria for each sampling criterion.
Again, the results show that missing matches randomly affect the distance calculation.
Figure 5-24 shows a clear trend where error decreases as the sampling criterion changes from 2.0
s to 8.0 s. There were no outliers seen for this specific DBS and the range of errors that occurred
from missing matches were very similar to the range of errors in the original experiment.
93
)%(
rorrE
egatnecreP
|
Colorado School of Mines
|
5.1.3.2 White-painted Wall
Figures 5-25 and 5-26 show the effect of missing matches for the white-painted wall
experiment when the DBS is 11.0 cm. The results presented in Figure 5-25 include all the
sampling criteria tested for each category of the matches missed. The results presented in Figure
5-26 include all the missing matches criteria for each sampling criterion.
Figure 5-25 shows that missing matches did not significantly affect the average
percentage error and did not cause major deviations from the normal calculation. However, the
misses did cause the ranges of errors to increase. Figure 5-26 shows that a sampling criterion of
one image every 8.0 s had the lowest accuracy when matches were missed.
25
20
15
10
5 Normal Calculation
Missing 1 match every other match
0 Missing 1 match every 2 matches
Missing 1 match every 5 Matches
-5 Missing 1 match every 10 Matches
-10
-15
-20
-25
Figure 5-25 Missing matches test - the effect of missing matches for the white-painted wall
experiment when the distance between the sensor is 11.0 cm.
95
)%(
rorrE
egatnecreP
|
Colorado School of Mines
|
25
20
15
10
5 Sampling at 1 image every 2 seconds
Sampling at 1 image every 4 seconds
0 Sampling at 1 image every 6 seconds
Sampling at 1 image every 8 seconds
-5 Sampling at 1 image every 10 seconds
-10
-15
-20
-25
Figure 5-26 Missing matches test - the effect of sampling criterion and matches missed for the
white-painted wall experiment when the distance between the sensor is 11.0 cm.
Figures 5-27 and 5-28 show the effect of missing matches for the white-painted wall
experiment when the DBS is 15.0 cm. The results presented in Figure 5-27 include all the
sampling criteria tested for each category of the matches missed. The results presented in Figure
5-28 include all the missing matches criteria for each sampling criterion.
Figure 5-27 shows that the normal calculation already had an outlier that overcalculated
the distance by 13%. This outlier remained the same when the misses were set to missing every
10 matches. The outlier then decreases in value when the misses were set to missing one match
every other match. Interestingly, an outlier that undercalculates distance by approximately 84%
appears when the algorithm misses every five matches. Figure 5-27 shows that this outlier
occurred when the sampling criterion was set to 10.0 s. Finally, misses are introduced into the
system, the average percentage error randomly changes.
The results show similar behavior to what we observed in the granite rock experiment,
where missing matches have a random effect on the accuracy of the distance calculation.
96
)%(
rorrE
egatnecreP
|
Colorado School of Mines
|
Figures 5-29 and 5-30 show the effect of missing matches for the white-painted wall
experiment when the DBS is 22.0 cm. The results presented in Figure 5-29 include all the
sampling criteria tested for each category of the matches missed. The results shown in Figure 5-
30 include all the missing matches criteria for each sampling criterion.
Figure 5-29 shows that the average percentage error did not significantly change when
misses were introduced in the system. However, the range of errors was affected by the misses.
Additionally, Figure 5-30 shows that sampling every 8.0 s resulted in the lowest accuracy for
distance calculation. Finally, an interesting observation is that sampling every 10.0 s had the
highest accuracy when misses were introduced. This was the only time that decreasing the
resolution of a sampling criterion increased the accuracy. Those are unexpected results, and as
discussed previously, it typically has the opposite effect on accuracy.
25
20
15
10
5 Normal Calculation
Missing 1 match every other match
0 Missing 1 match every 2 matches
Missing 1 match every 5 Matches
-5 Missing 1 match every 10 Matches
-10
-15
-20
-25
Figure 5-29 Missing matches test - the effect of missing matches for the white-painted wall
experiment when the distance between the sensor is 22.0 cm.
98
)%(
rorrE
egatnecreP
|
Colorado School of Mines
|
25
20
15
10
5
Sampling at 1 image every 2 seconds
Sampling at 1 image every 4 seconds
0 Sampling at 1 image every 6 seconds
Sampling at 1 image every 8 seconds
Sampling at 1 image every 10 seconds
-5
-10
-15
-20
-25
Figure 5-30 Missing matches test - the effect of missing matches for the white-painted wall
experiment when the distance between the sensor is 22.0 cm.
5.2 Experiment 2 – Optimizing Controllable Variables
This section discusses the results obtained from Experiment 2. According to the results
generated from Experiment 1, varying the sampling criterion while keeping the DBS constant
affects the accuracy of the distance calculation. Similarly, changing the DBS while keeping the
sampling criterion constant affects the accuracy of the distance calculation. Therefore, this
experiment aims to find an optimized combination of sampling criterion and DBS for different
velocities to minimize the errors especially those resulting from missed matches, false positives,
or changing wellbore conditions. As a result, distance calculation errors will be minimized. The
different average velocities tested were presented previously in Section 4.2.3.2. A sample of the
algorithm(cid:182)s velocity calculations for the first run when the average velocity of the cart is 63
ft/hour is shown in Tables A-1 to A-3 in the Appendix.
99
)%(
rorrE
egatnecreP
|
Colorado School of Mines
|
5.2.1 Results
Tables 5-7 and 5-8 show the average velocity and percentage error (explained in Section
4.2.4.1), respectively, calculated by the algorithm for each combination of sampling criterion and
DBS when the average velocity of the dolly is 63 ft/hour. Table A-4 shows the number of data
points used to calculate the average velocity computed by the algorithm. The results show that
the absolute value of each error was less than 5% for most of the combinations tested. However,
when the DBS is 30 cm, and the sampling is set to 20.0 s or 30.0 s, the errors were considerably
higher than the other combinations. This behavior is also observed for a DBS 45.0 cm and a
sampling criterion of 30.0 s. For a cart velocity of 63 ft/hour, a DBS of 15.0 cm provides
absolute errors lower than 10% for all sampling criteria tested.
Table 5-7 Average velocity calculated by the algorithm for the two 5-minute runs when the
average velocity of the dolly is 63 ft/hour.
Sampling Criteria 15.0 cm 30.0 cm 45.0 cm
2.0 s 60.5 ft/hour 63.7 ft/hour 63.2 ft/hour
5.0 s 61.4 ft/hour 62.8 ft/hour 62.6 ft/hour
12.0 s 58.1 ft/hour 59.3 ft/hour 61.4 ft/hour
20.0 s 62.0 ft/hour 54.6 ft/hour 61.6 ft/hour
30.0 s 59.0 ft/hour 51.3 ft/hour 54.2 ft/hour
100
|
Colorado School of Mines
|
Table 5-8 Percentage error resulting from the velocity calculated for the two 5-minute runs and
when the average velocity of the dolly is 63 ft/hour.
Sampling Criteria 15.0 cm 30.0 cm 45.0 cm
2.0 s -3.8% 1.2% 0.5%
5.0 s -2.5% -0.3% -0.5%
12.0 s -7.7% -5.7% -2.4%
20.0 s -1.5% -13.2% -2.2%
30.0 s -6.3% -18.4% -13.8%
Tables 5-9 and 5-10 show the average velocity and percentage error, respectively,
calculated by the algorithm for each combination of sampling criterion and DBS when the
average velocity of the dolly is 98 ft/hour. Table A-5 shows the number of data points used to
calculate the average velocity computed by the algorithm. The results show similar behavior to
Tables 5-7 and 5-8 when sampling at 2.0 s and 5.0s. The absolute errors occurring at those
sampling criteria are lower than 5% for all DBS. However, the absolute errors were higher for
the other combinations. Additionally, sampling every 30.0 s and using a DBS of 15.0 cm or 30.0
cm causes the algorithm to miss matches. Therefore, this sampling criterion should not be used
when the velocity of the cart is 98 ft/hour or higher because very high errors will be introduced
in the system.
101
|
Colorado School of Mines
|
Table 5-9 Average velocity calculated by the algorithm for the two 5-minute runs when the
average velocity of the dolly is 98 ft/hour.
Sampling Criteria 15.0 cm 30.0 cm 45.0 cm
2.0 s 101.1 ft/hour 101.3 ft/hour 96.6 ft/hour
5.0 s 100.9 ft/hour 93.66 ft/hour 95.3 ft/hour
12.0 s 108.6 ft/hour 80.3 ft/hour 83.1 ft/hour
20.0 s 72.8 ft/hour 78.5 ft/hour 81.6 ft/hour
30.0 s Misses Misses 79.0 ft/hour
Table 5-10 Percentage error resulting from the velocity calculated for the two 5-minute runs and
when the average velocity of the dolly is 98 ft/hour.
Sampling Criteria 15.0 cm 30.0 cm 45.0 cm
2.0 s 3.1% +3.4% -1.3%
5.0 s 3.0% -4.3% -2.7%
12.0 s 10.8% -17.2% -15.2%
20.0 s -25.8% -19.9% -16.7%
30.0 s Misses Misses -19.4%
Tables 5-8 and 5-10 show that the absolute errors that occur when sampling every 2.0 s
or 5.0s for all the distances between the sensors are lower than 5.0% for velocities ranging
between 63 ft/hour and 98 ft/hour. Therefore, it is a reasonable assumption that match misses
resulting from the tool(cid:182)s velocity will not occur if any of those combinations are picked. As a
result, to develop a simple model, minimize computational complexity, and mitigate chances of
false positives, the algorithm should sample every 5.0 s using a DBS of 15.0 cm.
Tables 5-11 and 5-12 show the average velocity and percentage error, respectively,
calculated by the algorithm for each combination of sampling criterion and DBS when the
average velocity of the dolly is 257 ft/hour. Table A-5 shows the number of data points used to
calculate the average velocity computed by the algorithm. The results show completely different
102
|
Colorado School of Mines
|
behavior than the model assumed for velocities ranging between 68 ft/hour and 98 ft/hour. The
absolute value of every combination that generates matches using a 15.0 cm DBS is higher than
40%. Additionally, the algorithm missed matches when sampling every 12.0 s, 20.0 s, or 30.0 s
using a 15.0 cm DBS. The results indicate that the model assumed for the previous velocities will
not accurately work for this velocity. Finally, Table 5-12 shows that the lowest errors for this
velocity occur when sampling every 2.0 s using a DBS of 30.0 cm or 45.0 cm or when sampling
every 5.0 s using a DBS of 45.0 cm.
Table 5-11 Average velocity calculated by the algorithm for the two 5-minute runs when the
average velocity of the dolly is 257 ft/hour.
Sampling Criteria 15 cm 30 cm 45 cm
2.0 s 152.4 ft/hour 245 ft/hour 255 ft/hour
5.0 s 138.5 ft/hour 216.8 ft/hour 246.4 ft/hour
12.0 s Misses Misses 191.7 ft/hour
20.0 s Misses Misses Misses
30.0 s Misses Misses Misses
Table 5-12 Percentage error resulting from the velocity calculated for the two 5-minute runs and
when the average velocity of the dolly is 257 ft/hour.
Sampling Criteria 15 cm 30 cm 45 cm
2.0 s -40.6% -4.3% -3.7%
5.0 s -46.0% -15.6% -4.0%
12.0 s Misses Misses -25.4%
20.0 s Misses Misses Misses
30.0 s Misses Misses Misses
103
|
Colorado School of Mines
|
Tables 5-13 and 5-14 show the average velocity and percentage error, respectively,
calculated by the algorithm for each combination of sampling criterion and DBS when the
average velocity of the dolly is 454 ft/hour. Table A-6 shows the number of data points used to
calculate the average velocity computed by the algorithm. The results show that when the
average velocity increases from 257 ft/hour to 454 ft/hour, sampling every 12.0 s or more will
cause the algorithm to miss matches. Table 5-13 shows that the lowest errors for this velocity
occur when sampling every 2.0 s using a DBS of 30.0 cm or 45.0 cm. Therefore, based on those
results and developing a simple model, it is a reasonable assumption that if the cart moves at a
velocity between 257 ft/hour and 454 ft/hour, the optimum combination should be sampling
every 2.0 s and picking a DBS of 30.0 cm. A DBS of 45.0 cm is unnecessary because it is more
computationally expensive than a DBS of 30.0 cm. A longer distance means that the algorithm
will have to match more images, but the accuracy will not significantly improve.
Table 5-13 Average velocity calculated by the algorithm for the two 5-minute runs when the
average velocity of the dolly is 454 ft/hour.
Sampling Criteria 15 cm 30 cm 45 cm
2.0 s 244.3 ft/hour 435.5 ft/hour 437.0 ft/hour
5.0 s Misses 308.6 ft/hour 317.8 ft/hour
12.0 s Misses Misses Misses
20.0 s Misses Misses Misses
30.0 s Misses Misses Misses
104
|
Colorado School of Mines
|
Table 5-14 Percentage error resulting from the velocity calculated for the two 5-minute runs and
when the average velocity of the dolly is 454 ft/hour.
Sampling Criteria 15 cm 30 cm 45 cm
2.0 s -46.1% -4.0% -3.5%
5.0 s Misses -32.0% -29.9%
12.0 s Misses Misses Misses
20.0 s Misses Misses Misses
30.0 s Misses Misses Misses
5.2.2 Optimized Combinations of Sampling Criterion and Distance between the Sensors
According to the results presented in Section 5.2.1, if the dolly moves within a velocity
range of 63 ft/hour to 98 ft/hour, the optimum combination is to sample every 5.0 s with a DBS
of 15 cm. Additionally, if the dolly moves within a velocity range of 257 ft/hour to 454 ft/hour,
the optimum combination is to sample every 2.0 s with a DBS of 30 cm. An optimum
combination of image sampling and DBS is not known for velocities between 98 ft/hour and 257
ft/hour. Ideally, testing more velocities within this velocity range would allow finding an
optimum combination. However, this work was limited to the velocities tested due to the RPM
settings available.
Experimental observation (Tables 5-7 to 5-14) determined that higher image sampling
frequency and longer DBS are required as velocity increases. There were missing image matches
when the velocity increased, and the algorithm was not sampling enough images.
It is evident from Section 5.2.1 that the optimum parameters must change somewhere
between 98 ft/hour and 257 ft/hour. Therefore, for velocities from 60 ft/hour to 120 ft/hour, the
optimum combination is to sample every 5.0 s with a DBS of 15.0 cm. Then for velocities
105
|
Colorado School of Mines
|
greater than 120 ft/hour and lower than 455 ft/hour, the optimum combination used is sampling
every 2.0 s with a DBS of 30.0 cm (Table 5-15).
Table 5-15 Optimized combinations of sampling criterion and distance between the sensors for
different velocity ranges.
Velocity Range
Optimum Distance between the
Optimum Sampling Criterion
(ft/hour) Sensors (cm)
60 – 120 1 image/5.0 s 15.0
120 – 454 1 image/2.0 s 30.0
5.3 Experiment 3 – Real-time Distance Calculation on Actual Well Logs
In this experiment, the model presented in Table 5-15 is added to the image-matching
algorithm. The experiments are then performed in real-time on actual ultrasonic well logs. The
algorithm first calculates the velocity and then picks the optimum combination of sampling
criterion and DBS based on the velocity calculated. The objective of this experiment is to
observe how the modified algorithm will behave and if the distance calculation errors are
minimized. A total of three different tests were made for this experiment, as demonstrated earlier
in Section 4.2.5.
Figure 5-31 shows the distance calculations and cumulative error from the first test,
where the dolly was roughly moving at a constant velocity. The figure shows that the optimized
combination caused the distance calculated to be close to the actual distance measured by the
tape. The distance was overcalculated from 10 s to 100 s and undercalculated from 120.0 s to
520.0 s. The total cumulative error for the whole 13.5 ft experiment is -0.3 ft. It is expected that
optimized combinations will never be error-free since errors are always encountered when
106
|
Colorado School of Mines
|
0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
-0.12
-0.14
Figure 5-32 Error distribution of Test 1 based on 100 data points.
Figure 5-33 shows the distance calculations and cumulative error from the second test,
where the dolly(cid:182)s velocity was randomly changing. The figure shows that the optimized
algorithm(cid:182)s calculation follows the trend of the actual distance traveled. However, there were
times when the distance calculated by the algorithm and the actual distance traveled diverged, for
example, from 170.0 s to 220.0 s. The total cumulative error for the whole 13.5 ft experiment is -
0.8 ft. Another interesting observation is that the velocity of the cart shown previously in Figure
4-23 and the cumulative error presented in Figure 5-39 are inversely correlated. This relationship
is demonstrated more clearly in Figure 5-34. The figure shows that as the velocity increases, the
cumulative error shifts towards a negative value, and when the velocity decreases, the
cumulative error shifts towards a positive value. This observation is expected because, for low
velocities, the optimum combination has positive errors, and for high velocities, it has negative
errors.
108
)tf(
rorrE
|
Colorado School of Mines
|
Figure 5-35 shows the distribution of errors for Test 2 using a box and whiskers plot. The
plot consists of 90 calculation points. The figure shows that the mean error is 0.01 ft.
Additionally, the standard deviation is 0.07 ft and the maximum error encountered is -0.25 ft.
The plot is more dominated by positive errors rather than negative errors.
0.25
0.2
0.15
0.1
0.05
0
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
Figure 5-35 Error distribution of Test 2 based on 90 data points.
Figure 5-36 shows the distance calculations and cumulative error from the third test,
where the dolly(cid:182)s velocity was randomly changing. The figure shows that the optimized
algorithm(cid:182)s calculation follows the trend of the actual distance traveled. Additionally, the
calculations performed by the algorithm are more similar to the actual distance traveled when
compared to Test 2. However, there were still times when the distance calculated by the
algorithm and the actual distance traveled diverged, such as 50.0 s to 120.0 s. The total
cumulative error for the whole 13.5 ft experiment is 0.0 ft. Again, this test demonstrates similar
110
)tf(
rorrE
|
Colorado School of Mines
|
CHAPTER 6
DISCUSSION
This section discusses and summarizes the results of all the experiments performed in this
dissertation. Additionally, it discusses this work(cid:182)s contribution towards directional drilling
automation and the potential downsides and challenges of the proposed method.
6.1 Experiment 1 – Image-Matching Code Accuracy and Key Variables
The first experiment was performed to address the research questions posed at the end of
Section 2.1. The aim of the experiment was to discover if topological features combined with
SIFT, FLANN, and GRT can be used to compute the distance traveled. Therefore, the accuracy
of the matches performed by the algorithm was tested using sceneries, such as granite rock and a
white-painted wall. Additionally, the experiment also tested the effect of controllable and
uncontrollable variables on the distance calculation. Parameters such as sampling, distances
between the sensors, matches missed, and noise were all tested to see how the accuracy of the
distance calculation changes.
6.1.1 General Image-Matching using Granite Rock and White-Painted Wall
Every similar match that the algorithm computed between two images was visually
inspected for confirmation. This inspection demonstrated that the algorithm matches similar
images correctly and assigned them the correct timestamps. The results indicate that if imaging
sensors capture apparent topological features, then using SIFT, FLANN, and GRT will allow to
determine a match between images captured at different sensors.
114
|
Colorado School of Mines
|
6.1.2 The Effect of Sampling and the Distance Between the Sensors
For the granite rock test, Figure 6-1 summarizes how changing the DBS affects the
accuracy of the distance calculated. The results presented for each DBS include all the sampling
criteria combined. For example, the range of percentage errors seen for a DBS of 8.3 cm consists
of the results obtained from sampling at one image every 2.0 s, 4.0 s, 6.0 s, and 8.0 s.
Figure 6-1 shows that as the DBS increases, the average percentage error shifts from a
negative to a positive value. The range of percentage errors demonstrated by a DBS of 15.0 cm
suggests that the error can vary significantly when the sampling criterion is changed. Another
interesting observation is that the errors for all distances between the sensors can be positive or
negative.
20
15
10
Distance between sensors = 8.3
5 cm
Distance between sensors = 13.3
0
cm
-5 Distance between sensors = 15
cm
-10
-15
-20
Figure 6-1 DBS test - summary of the results obtained from the granite rock test for all the
sampling criteria at each distance between the sensors.
For the granite rock test, Figure 6-2 summarizes how changing the sampling criterion
affects the accuracy of the distance calculated. The results presented for each sampling criterion
include all the DBS combined. For example, the range of percentage errors seen for a sampling
criterion 2.0 s consists of all the results obtained from a DBS of 8.3 cm, 13.3 cm, and 15.0 cm.
115
)%(
rorrE
egatnecreP
|
Colorado School of Mines
|
Figure 6-2 shows that the range of errors for each sampling criterion tested can be
positive or negative. The range of errors also varies as the sampling changes, suggesting the
strong effect sampling has on the image-matching algorithm. Additionally, the range of errors for
a sampling every 2.0 s is wide, implying that the errors can also significantly vary when the DBS
changes.
20
15
Sampling Criterion = 1 Image/2
10
seconds
5 Sampling Criterion = 1 Image/4
seconds
0
Sampling Criterion = 1 Image/6
-5 seconds
Sampling Criterion = 1 Image/8
-10
seconds
-15
-20
Figure 6-2 DBS test - summary of the results obtained from the granite rock test for all the
distances between the sensors at each sampling criterion.
For the white-painted wall test, Figure 6-3 summarizes how changing the DBS affects the
accuracy of the distance calculated. The results presented for each DBS include all the sampling
criteria combined. For example, the range of percentage errors observed for a DBS of 11.0 cm
includes the results obtained from sampling at one image every 2.0 s, 4.0 s, 6.0 s, 8.0 s, and 10.0
s.
Figure 6-3 shows that as the DBS increases, the average percentage error shifts from a
positive to a negative value. This is the exact opposite of the trend observed in Figure 6-1 for the
granite rock test. Additionally, the errors for a DBS of 11.0 cm and a DBS of 15.0 cm were
positive or negative values. However, when using a DBS of 22.0 cm, all errors were negative,
and the distance was always undercalculated.
116
)%(
rorrE
egatnecreP
|
Colorado School of Mines
|
20
15
10
5
Distance between sensors = 11 cm
0 Distance between sensors = 15 cm
Distance between sensors = 22 cm
-5
-10
-15
-20
Figure 6-3 DBS test - summary of the results obtained from the white-painted wall test for all the
sampling criteria at each distance between the sensors.
For the white-painted wall test, Figure 6-4 summarizes how changing the sampling
criterion affects the accuracy of the distance calculated. The results presented for each sampling
criterion include all the DBS combined. For example, the range of percentage errors observed
when sampling at 2.0 s consists of the results obtained from a DBS of 11.0 cm, 15.0 cm, and
22.0 cm.
Figure 6-4 shows that the errors for each sampling criterion can be positive or negative.
Additionally, the range of errors when sampling at 8.0 s is wide, suggesting that the errors can
significantly vary when the DBS changes. Finally, when sampling every 10.0 s, the range of
errors is narrow, and the distance calculation is more accurate than other distances between the
sensors.
117
)%(
rorrE
egatnecreP
|
Colorado School of Mines
|
20
15
Sampling Criterion = 1 Image/2
10 seconds
Sampling Criterion = 1 Image/4
5
seconds
Sampling Criterion = 1 Image/6
0
seconds
Sampling Criterion = 1 Image/8
-5
seconds
-10 Sampling Criterion = 1 Image/10
seconds
-15
-20
Figure 6-4 DBS test - summary of the results obtained from the white-painted wall test for all the
distances between the sensors at each sampling criterion.
It is evident from the results presented in Figures 6-1 to 6-4 that a DBS combined with a
specific sampling criterion can significantly affect the accuracy of the distance calculation.
Those results imply that some combinations will be more accurate than others. It is, therefore,
imperative to consider an optimal combination of sampling and DBS when designing the image-
matching algorithm and the image-capturing tool.
6.1.3 The Effect of Noise and the Accuracy of Denoising
For the granite rock test, Figure 6-5 summarizes the accuracy of the distance calculated
for each type of test. The results presented for each type of test include all the sampling criteria
combined. For example, the range of percentage errors for the noisy test consists of the results
obtained from sampling at one image every 2.0 s, 4.0 s, 6.0 s, and 8.0 s.
Figure 6-5 shows that the range of errors resulting from the denoised test is similar to the
original test. In fact, the denoised test provides more accurate results and has a better average
percentage error. The results, therefore, indicate that the denoising algorithm can remove random
118
)%(
rorrE
egatnecreP
|
Colorado School of Mines
|
speckle noise and maintain the accuracy of the distance calculation. Finally, the noisy test had
the lowest accuracy out of all tests and undercalculates the distance.
25.00
20.00
15.00
10.00
5.00
Original
0.00 Noisy
Denoised
-5.00
-10.00
-15.00
-20.00
-25.00
Figure 6-5 Noise test - summary of results from the granite rock test.
For the white-painted wall test, Figure 6-6 summarizes the accuracy of the distance
calculation for each test type. The results presented for each type of test include all the sampling
criteria combined. For example, the range of percentage errors seen for the noisy test consists of
the results obtained from sampling at one image every 2.0 s, 4.0 s, 6.0 s, 8.0 s, and 10.0 s.
The results are slightly different for this test compared to the granite rock test. The errors
resulting from the denoised test did not entirely fall within the range of errors that resulted from
the original test. Some of the sampling criteria caused the denoising algorithm to overcalculate
the distance and have higher errors than those observed in the original experiment. However,
most of the errors from the denoised test are within the range of the original test, and the average
percentage error from both tests is similar. Finally, the average percentage error from the noisy
test is the most accurate out of all tests. However, the noisy test seems to often undercalculate the
distance. Again, the results emphasize that noise can cause the distance calculation to behave
119
)%(
rorrE
egatnecreP
|
Colorado School of Mines
|
6.1.4 The Effect of Missing Matches
The results presented in Section 5.1.3 show that missing matches can harm the accuracy
of the distance calculation. It occurs when the image-matching algorithm cannot find similar
image pairs to match, leading the algorithm to skip that image pair and move on to the next. As a
result, the image-matching algorithm cannot compute the tool's velocity at the time of the misses.
It assumes that the velocity of the missed timestamp is the same as the previous velocity
calculated. If the assumption is correct, then the distance calculation will not be highly affected;
however, if it is incorrect, then serious random errors are introduced in the system. There are
four main reasons why misses occur:
1. The tool is going at a high velocity, and the sampling rate combined with the DBS is
not optimized in a way that allows the image-matching algorithm to find similar
images.
2. The sensors have a limited resolution and do not adequately capture the topological
features of the borehole wall required for the image-matching process.
3. Noise is constantly changing the image(cid:182)s features.
4. The wellbore conditions are constantly changing, causing the images captured by
identical sensors also to change.
A possible solution for mitigating misses due to high tool velocity is to optimize
sampling and DBS for different velocity ranges, as done in this work in Experiment 2.
Additionally, if the sensor resolution is affecting the image(cid:182)s features, then an algorithm that
reconstructs the surface topology may be required to mitigate misses occurring from the lack of
features in images. The downside of this additional algorithm would be the extra computational
time required. Finally, to mitigate misses occurring due to wellbore conditions, an optimal mud
121
|
Colorado School of Mines
|
design should be considered before implementing the tool in field applications. Drilling fluids
control the integrity of the borehole, and a well-designed mud may prevent downhole conditions
from constantly changing. Additionally, the sensors should be as close to the bit as possible to
prevent changes in wellbore conditions from affecting the images. Theoretically, the more time
the sensors wait until they capture an image, the higher the risk of wellbore conditions changing.
6.2 Experiment 2 – Optimizing Controllable Variables
The results presented in Section 5.2.1 showed that a single combination of sampling
criterion and DBS would have different velocity calculation accuracies depending on the actual
velocity of the tool. Since the distance traveled is a function of the calculated velocities, the
distance calculation will carry over those errors. Table 6-1 summarizes the results for all
combinations made using a DBS of 15.0 cm. The table shows that every combination has errors,
but some are much higher than others. It can be seen that as velocity increases, the absolute
errors using a DBS of 15.0 cm increase. Additionally, as the number of samples captured
decreases, the absolute errors will, in general, increase.
Table 6-1 Velocity calculation errors when using a DBS of 15.0 cm.
Combination of Sampling
63 ft/hour 98 ft/hour 257 ft/hour 454 ft/hour
Criterion and DBS
2.0 s and 15.0 cm -3.80% 3.10% -40.60% -46.10%
5.0 s and 15.0 cm -2.50% 3.00% -46.00% Misses
12.0 s and 15.0 cm -7.70% 10.80% Misses Misses
20.0 s and 15.0 cm -1.50% -25.80% Misses Misses
30.0 s and 15.0 cm -6.30% Misses Misses Misses
122
|
Colorado School of Mines
|
Table 6-2 summarizes the results for all combinations made using a DBS of 30.0 cm.
Similar to Table 6-1, it can be seen that as velocity increases, the absolute errors using a DBS of
30.0 cm increase. Additionally, when the number of samples captured decreases, the absolute
errors increase. However, the absolute errors for the velocities higher than 257 ft/hour are much
lower when a DBS of 30.0 cm is used instead of 15.0 cm. This observation suggests that the
sampling rate or the DBS should increase as velocity increases to prevent misses and minimize
errors.
Table 6-2 Velocity calculation errors when using a DBS of 30.0 cm.
Combination of Sampling
63 ft/hour 98 ft/hour 257 ft/hour 454 ft/hour
Criterion and DBS
2.0 s and 30.0 cm 1.20% 3.40% -4.30% -4.00%
5.0 s and 30.0 cm -0.30% -4.30% -15.60% -32.00%
12.0 s and 30.0 cm -5.70% -17.20% Misses Misses
20.0 s and 30.0 cm -13.20% -19.90% Misses Misses
30.0 s and 30.0 cm -18.40% Misses Misses Misses
Table 6-3 summarizes the results for all combinations made using a DBS of 45.0 cm.
Similar to Tables 6-1 and 6-2, it can be seen that as velocity increases, the absolute errors using a
DBS of 30.0 cm increase. Also, when the number of samples captured decreases, the absolute
errors increase. When comparing Table 6-2 with Table 6-3, it can be seen that the absolute errors
for all velocities are lower when a DBS of 45.0 cm is used instead of 30.0 cm. However, the
differences are insignificant when sampling every 2.0 s or 5.0 s.
123
|
Colorado School of Mines
|
Table 6-3 Velocity calculation errors when using a DBS of 45.0 cm.
Combination of Sampling
63 ft/hour 98 ft/hour 257 ft/hour 454 ft/hour
Criterion and DBS
2.0 s and 45.0 cm 0.50% -1.30% -3.70% -3.50%
5.0 s and 45.0 cm -0.50% -2.70% -4.00% -29.90%
12.0 s and 45.0 cm -2.40% -15.20% -25.40% Misses
20.0 s and 45.0 cm -2.20% -16.70% Misses Misses
30.0 s and 45.0 cm -13.80% -19.40% Misses Misses
Based on the results observed in Tables 6-1, 6-2, and 6-3, this work developed two
simple models that cause the algorithm to switch between optimized combinations based on the
tool(cid:182)s velocity. It can be seen from Table 6-1 that if the cart is moving at a velocity that ranges
from 63 ft/hour to 98 ft/hour, the algorithm should sample every 5.0 s and use a DBS of 15.0 cm.
Other combinations are also possible but unnecessary because this combination provides less
than 5% absolute errors and decreases computational time compared to combinations that use a
sampling criterion of 2.0 s. Additionally, errors may occur due to false positives if the DBS is
longer than needed because more images are being matched. Similarly, based on Table 6-2, if the
cart is moving at a velocity that ranges between 257 ft/hour to 454 ft/hour, the algorithm should
sample every 2.0 s and use a DBS of 30.0 cm.
The optimal combinations of the velocities ranging between 98 ft/hour to 257 ft/hour
were unknown. Based on the optimal combinations for the other velocities and the apparent shift
between models as the cart velocity increases (Table 6-4), it was assumed that the two models
would remain the same for the unknown velocities and that the shift between the models would
occur somewhere in the unknown region. Model 1 used a DBS of 15.0 cm and a sampling
criterion of 5.0 s for velocities ranging from 60 ft/hour to 120 ft/hour. Model 2 used a DBS of
30.0 cm and a sampling criterion of 2.0 s for velocities between 120 ft/hour and 454 ft/hour.
124
|
Colorado School of Mines
|
Table 6-4 Velocity calculation errors for the optimized models
Combination of Sampling
63 ft/hour 98 ft/hour 257 ft/hour 454 ft/hour
Criterion and DBS
5.0 s and 15.0 cm -2.50% 3.00% -46.00% Misses
2.0 s and 30.0 cm 0.50% -1.30% -3.70% -3.50%
As demonstrated by the models, the cut-off between the two models was chosen to be
120 ft/hour. This simple assumption was based on the theory that as the velocity increases,
sampling, DBS, or both should also be changed and optimized to prevent misses. The results
indicate that a combination of 5.0 s and 15.0 cm will cause high errors at high velocities.
Therefore, to prevent a rapid increase in errors, a higher range of velocities was chosen to follow
Model 2 instead of Model 1 since a combination of 2.0 s using a DBS of 30.0 cm provides low
errors at all velocities.
It is important to note that this cut-off may not have been the appropriate value. A safer
assumption could have been to choose 2.0 s and 30.0 cm for all velocities higher than 98 ft/hour.
However, this cut-off was chosen to decrease the required number of samples and thus decrease
computational complexity. The results need to appear as fast as possible in real-time. If
processing time is high, it is very difficult for the tool to make continuous autonomous decisions.
Nevertheless, the user has the option to choose whichever combination that fits his desired
accuracy. The tests made here only show the methodology but not necessarily the most accurate
combination.
The most important observations here is that the optimized combinations used to
calculate the velocity are within ± 5% errors (similar to the pipe tally), suggesting that the
distance calculation will most likely not have higher than ±5% errors. However, errors may have
fluctuated from ± 5% for the untested velocities. Testing more velocities, adding more sensors,
and testing a more comprehensive range of sampling criteria may help in finding better
125
|
Colorado School of Mines
|
combinations than the optimized combinations chosen. Additionally, there could have been a
better combination that provides similar accuracies but at a lower computational cost.
Nevertheless, while this work is limited to only the combinations tested, the same methodology
can be applied to a wider range of combinations and velocities, thus optimizing the system even
further, and possibly enhancing accuracy.
6.3 Experiment 3 - Real-time Distance Calculation on Actual Well Logs
Experiment 3 tested the optimized model in real-time on actual ultrasonic logs. There
were many important observations that the tests showed:
1. The actual distance traveled and the distance calculated by the algorithm were similar
and followed the same trend with some errors encountered. The similarity between
curves proves that using identical imaging sensors and matching similar images to
compute the distance can work. More work can be done to optimize the system
further and enhance the accuracy of the measurements.
2. Figure 6-7 shows the distribution of errors combined from all the experiments made.
It is evident that the errors roughly have a zero mean. The maximum errors
encountered on both sides of the box are + 0.25 ft and – 0.25 ft. Future work should
aim to find a way for errors to remain between the first and third quartile, where the
errors are maintained close to zero.
126
|
Colorado School of Mines
|
CHAPTER 7
CONCLUSION AND RECOMMENDATIONS FOR FUTURE WORK
This chapter concludes the work presented in this dissertation and recommends different
tests that could be made in the future to improve the proposed method.
7.1 Conclusion
Bottomhole distance measurements have been a significant constraint to the automation
of directional drilling operations. As demonstrated in the literature, bottomhole measurement
techniques proposed today have operational limitations and high calculation errors. For example,
the geological correlation techniques proposed by multiple researchers can be highly inaccurate
in homogeneous sections due to a lack of change in rock properties. Therefore, this research aims
to provide the industry with a sound, robust, and accurate method for measuring the distance
drilled at the bottom of the well. The proposed approach can be integrated with a rotary steerable
system where constant measurements of distance, inclination, and azimuth allow the downhole
computer to continuously locate the bit and change its location accordingly.
The method proposed here is not designed to fully disconnect from surface
measurements, but designed to allow long periods of automated drilling. If the tool cannot
measure the drilled distance due to severe bottomhole conditions, the directional driller can
control until conditions improve. Additionally, if the tool cannot match the images due to a lack
of rock features, errors will occur from the measurement, and surface intervention and
corrections will be required. The goal is to have this tool work under most conditions, have
relatively low errors, and limit the need for fine-tuning and surface dependence.
129
|
Colorado School of Mines
|
The proposed method consists of multiple identical imaging sensors located at known
positions, collecting similar fingerprints of the formation at different but synchronized times. A
key-points identification algorithm identifies the fingerprints, which are a continuous
representation of the natural geological features of the rock, the formation structure, and the
marks left by the drilling bit. An image-matching algorithm compares the fingerprints of the
images and calculates a percentage similarity for all comparisons. A cubic interpolation between
the percentage similarities identifies the approximate time that two similar images matched. The
process then uses the timestamps and the distances between the sensors to determine the
velocity. Finally, the algorithm integrates the velocity to compute the interval distance drilled.
Continuously adding the interval distance drilled to the previous total will result in the current
MD.
Three different types of experiments were performed to verify the efficacy of the
proposed method. The experiments were performed on natural sceneries, including granite rock,
a white-painted wall, and a gravel road. Additionally, ultrasonic logs were tested as a final
evaluation for the proposed method. The first experiment was executed manually, while the
second and third experiments used a speed-adjustable motorized cart.
The first experiment tested the effect of controllable and uncontrollable variables on the
accuracy of the distance calculation. This experiment showed that sampling and sensor location
significantly affect the accuracy of the distance calculation. Additionally, random noise causes
errors and requires an efficient denoising algorithm that retains the key features lost from the
noisy image. Finally, missing matches could cause random and high errors to occur.
The second experiment aimed to optimize key variables and minimize calculation errors
occurring from match misses. The results show that the sampling rate and the distance between
130
|
Colorado School of Mines
|
the sensors should be changed when the velocity changes. If the tool moves at a high velocity
and not enough images are captured, the system starts missing matches, and the accuracy
decreases. Similarly, if the tool moves at a high velocity, and the sensors are located at a very
short distance from each other, misses are also introduced in the system, and again, the accuracy
decreases. As a result of the observations, the algorithm was modified to capture the images
using an optimized combination of sampling and distance between the sensors for different
velocities.
The third experiment tested the modified image-matching algorithm in real-time using
ultrasonic logs. This experiment showed that misses could be mitigated when the sampling and
the distance between the sensors are optimized. The results also showed similar trends between
the distance calculated by the algorithm and the actual distance traveled, suggesting that an
optimized combination can be used to minimize errors.
This work proves that the proposed method can measure the distance traveled. However,
more work is required to control the errors and minimize them. Throughout the testing process, it
was discovered that the errors are not random, can be correlated with the velocity, and follow a
Gaussian distribution. As a result, more work needs to be done to increase the accuracy of the
system.
7.2 Recommendations for Future Work
The work presented in this research proved an image-matching process that identifies
topological features can be used to estimate the distance traveled. However, more testing is
needed to enhance the algorithm(cid:182)s accuracy and transform this methodology from the lab to the
field. The limitations of the experiments accompanied by their future work recommendations
are:
131
|
Colorado School of Mines
|
1. Type of equipment and sceneries used for testing: In my experiments, I used a motorized
dolly with optical cameras to capture images of sceneries that imitate bottomhole
formations. I was limited to this equipment because an imaging while drilling tool with
similar sensors is not readily available. For future work, researchers should design and
develop an imaging tool with similar ultrasonic sensors and repeat the experiments in a
real wellbore, preferably a homogeneous section.
2. Computational power for real-time data analysis: I used a standard laptop to run real-time
experiments, and there was always a delay when getting the results back. For future
work, researchers should use a dedicated processor when testing. Additionally, they
should optimize the image-matching algorithm so that data is analyzed efficiently in real-
time.
3. Type of noise added and denoising methodology: I tested the effect of random speckle
noise on the image-matching algorithm and developed a denoising algorithm to restore
the images(cid:182) features. However, random speckle noise does not fully represent the noise
resulting from drilling operations. For future work, researchers should test the ultrasonic
imaging tool in a wellbore with circulated cuttings. Additionally, they should also
develop a more robust denoising algorithm that will remove the noise added to the
images from the cuttings and restore the images(cid:182) features.
4. The number of velocities tested and the optimized models developed: In my experiments,
I only tested four velocities to find the optimized combination of sampling criterion and
distance between the sensors for each velocity. I only used two models to keep the
algorithm simple. For future work, researchers should test more velocities and see how
changing the combination for each velocity increases the accuracy of the distance
132
|
Colorado School of Mines
|
ABSTRACT
The uncertainties in the energy and industry sector cascade coal demand uncertainties,
which challenge coal company's supply chain process. Hence understanding the relationship
between the planning, production, blending, transport, and sales process are crucial for
effective decision making. These relations are also vital to creates a business plan as a
decision support tool to define their abilities to meet the demand. The research presents
system dynamics modeling as decision support tool in coal business supply chain. The
proposed approach utilizes a coal company's production, transportation, blending, and sales
data combined with future scenarios simulation.
Actual data validate the model to ensure its accuracy. Illustrations about how much
coal producers should produce specific coal based on demand are presented in the model.
Inversely, the model also predicted the demand fulfilment based on company’s coal
production and availability. Based on the result and analysis, the transport model has reached
73% accuracy based on the Absolute Accuracy Calculation, RSME of 65,365.07 and 21%
error rate from MAPE. The stockpile component has 50.04% for absolute accuracy,
115,220.59 for RMSE calculation, and 49.95% for MAPE calculation.
Three scenarios are introduced into the model including, diversification scenario, 2022
supply-demand scenario, and 2023 supply-demand scenario. The component with higher
accuracy tends to have similar behavior in all the scenarios. Meanwhile, there are some
differences in the behavior in the component with lower accuracy. But in overall, with this
system dynamics model the company can increase their ability to creates more faster and
accurate production, sales, and future scenario planning with more precise calculation. This
happen because they have more detailed information about their supply chain condition, with
visualization of the future behavior.
iii
|
Colorado School of Mines
|
ACKNOWLEDGEMENT
In the name of Allah, the Merciful and Beneficent. All praise and appreciation are due
to Allah Almighty, Who has made everything possible through His unlimited generosity and
mercy. Shalawat and salaam are addressed to Muhammad PBUH, the noble prophet who has
led all of humanity from darkness to light, and from ignorance to wisdom.
First and foremost, I want to thank my advisor, Dr. H. Şebnem Düzgün, for her
guidance, constructive criticism, comments, and thesis ideas. Working with her has made this
difficult and fast-paced research a pleasurable part of my master's program. Throughout this
process, I am grateful for her support and encouragement.
I would like to show my thankfulness to Dr. Jürgen F. Brune and Dr. Tulay Flamand,
members of my thesis committee, for their advice, criticism, and suggestions. My heartfelt
appreciation also goes to Dr. Stephen M. Enders, the Mining Engineering program's director.
I want to express my sincere thanks to PT. Bukit Asam, my primary sponsor and
employer, for supporting my master's degree at Colorado School of Mines. Thank you to all
my Colorado friends from the Indonesia Students Association (PERMIAS) of Golden and
MINES for all your support, assistance, encouragement, inspiration, and shared experiences.
Go Orediggers!
I am grateful to my mother, fathers, and in-laws for their unending love, support,
passionate prayers, and encouragement. Kudos to my brothers and sisters for their inspiration
and support. Last but the very important ones, for their patience, prayers, and support
throughout this journey, I am eternally grateful to my lovely wife Annisaa Rahmi Said and
my adorable little dinosaurs Muhammad Karoobi Abghary Mirza.
xi
|
Colorado School of Mines
|
CHAPTER 1
INTRODUCTION
Since the 2nd century, humans have used coal as an energy source, and it was the
primary source of fuel for the Aztecs and Romans. The trend continued throughout the
industrial revolution, which started to utilize coal as a heating source, for steam engines and
as an electricity generator. Today, coal is still considered an essential energy source. More
than that, coal is also used for steel production, as well as other industrial, commercial, and
institutional uses (USGS, 2018). In fact, it is recorded that coal had total consumption of
7,906 million tonnes, rising 6% from its level of demand in 2020 (IEA, 2021). They also
predicted in 2021 that coal demand was rebounding for three main reasons: a recovering
global economy, a cold winter and hotter summer than 2020 increased power demand, and
supply issues that pushed gas prices higher (IEA, 2021). Many developed countries,
including China, the U.S., and India, are still racing to increase their coal demand despite
climate change organizations pushing to use more environmental-friendly energy. China still
considers coal as a crucial part of the energy mix in its long-term agenda (Forbes, 2021).
Figure 1.1 Global coal consumption by region, 2000-2024 (Source: IEA (2021) Coal 2021
Analysis and Forecast to 2024. All rights reserved.)
From Figure 1.1, The International Energy Agency (2021) stated that coal production
will expand 4.5% from 2020, resulting in total world production reaching 7.8 Mt. It believes
producers are still trying to improve their total production in future years (IEA, 2021). That is
a fact many countries are still dependent on coal for their electricity. As the least expensive
energy source currently available, coal is still the first choice in the energy business,
especially in developing countries.
1
|
Colorado School of Mines
|
As one of the developing countries and the third-largest coal-producing country
globally (IEA, 2021), Indonesia also predominantly depends on coal. Proved by one of the
Indonesian president’s goals to build 35,000 megawatts (MW) of power plants with coal as
primary energy sources by 2025. Kontan Magazine (2021) released an article about this
power plants project which, until August 2021, has reached 30% of its total target. Since
2019, the data also shows that coal has been the most prominent energy supply in Indonesia,
replacing crude oil and associated products that have dominated there for several decades
(ESDM, 2020). Ministry of Energy and Mineral Resources (2020) also stated that Indonesia
still has 38,805 Mt of total coal reserves with an annual production of 500-600 Mt per year.
These predictions and facts are a breath of fresh air for coal companies to increase production
and advance their respective businesses.
Coal as the most prominent energy has several different types with various calorific
values (CVs). These types include peat with the lowest CVs, lignite, sub-bituminous,
bituminous and anthracite which has the highest CVs on the heat coal. Coal’s natural forming
process with different temperatures and pressures is the agent which creates the difference.
Uniquely, all types are used to fulfill different needs, including power plants, factories, and
households. These varying types of coal can also lead to differences in reserve availability
and later affect product availability. Coal with higher quality will also have a higher price
than the lower ones. In most cases, higher quality coal can satisfy a lower quality
specification; however, the lower quality coal is not eligible to fulfill the requirements for
higher specification. Therefore, coal cannot be sold uniformly. Here is why there is a
blending process introduced to the coal business. There are many business situations when a
company cannot fulfill a customer’s original demand because of their limited reserves,
resulting in the company needs to blend their higher quality coal with the lower quality to
still fulfill the volume requirements and avoid penalty or contract under performance.
Although we can project the annual coal consumption of a country and company, there
are still times when there is sudden demand due to a lack of inventory. This situation could
arise for several reasons, including an expected shipment failure, a production delay, and
other supply chain problems. This sudden demand creates an opportunity for coal companies
to sell their available products of the same or similar specifications. These opportunities can
only be executed with fast decision-making from company’s management, with a knowledge
of its real-time stockpile condition and their probability product blending. Knowledge of
those information also indirectly assesses the entire supply chain process.
2
|
Colorado School of Mines
|
However, coal industry has a complex system for its production and sales because of
coal ranging qualities. Adapting coal from a mine-product quality to match a market-product
quality can be a challenge. All coal companies should have a systematic flow that ensures
their production planning will fulfill sales requirements. Therefore, a synergy between sales,
production and transport is needed to ensure no miscommunication within the process.
Understanding every process in the coal industry supply chain is crucial. The connection
between the upstream sector, including mine planning and production, the transportation and
blending processes, and then the downstream sector, namely sales; must be well established
to ensure accuracy of inventory at any time. Real-time inventory updates help company’s
management to make a business decision faster.
With a faster pace in all businesses, a decision support tool is necessary to create better
and faster decision-making in ensuring a business runs smoothly. The coal industry is no
exception. A decision support tool needs to have the ability to look at the whole coal business
situation comprehensively. It should have the capability to model the entire system and the
relationships between systems in detail. These tools also must be able to find the root of the
problem and predicting futures output by looking at behavior changes between the systems.
One way to solve this is by using system engineering principle, which helps the coal industry
tackle the problem, especially inventory management. With a wide variety of mine products
that can be generated into several sales’ product, it is essential to predict coal input and
output in different scenarios. With fast and accurate predictions, management can pinpoint
decisions to secure business opportunities. Therefore, this thesis aims to implement systems
engineering principles to create a decision support tool by using System Dynamics (SD) to
tackle Indonesian coal company’s supply chain problem.
SD is a computer-assisted strategy and policy formulation method that can re-create an
entire system via computer-aided software (System Dynamics Society, 2021). We can
visualize actual conditions and create a simulated result showing systems behavior by
creating a system dynamics model. The model can help businesses create faster and more
accurate decisions by generating scenarios that are likely to be faced by company’s
management, including those in the coal industry. This is done by adding new parameters to
reflect the probable future condition, for example, diversifying products or providing
simulations with several supply and demand conditions into the model. A literature
assessment of the above information is provided in the following subsections to establish the
thesis's motivation, goal, and contribution.
3
|
Colorado School of Mines
|
1.1 State of the Art
The approaches that are used to perform this research is discussed in the following
subsections. Starting with a description of the coal industry's supply chain and ending with an
overview to SD model. This chapter also discusses the fundamentals of SD modeling and the
factors to be considered in modeling the specific case study.
1.1.1 Mine-to-Market Supply Chain in the Coal industry
Almost every industry globally has several levels of supply chain processes. The
mining sector, particularly coal mining, has its own supply chain, a sequence mainly
composed of production planning, production, blending, transportation, and sales. Mine
planning is done in accordance with the resource model as well as the existing sales plan. The
planning process is followed by the production process, which is carried out in compliance
with existing mining standards. After that, the following process is the transport process. The
transport process consists of two different stages: getting the extracted coal to the stockpile
and transporting coal from the mine site to the location for sale.
As coal quality cannot be fully controlled by resource modeling due to involved
uncertainties, a blending process is introduced between two transport processes to obtain
specific coal specifications for the customers. Extra attention is needed in this blending
process to ensure that the entire distribution of coal is effective and in accordance with
customer demands. This process requires detailed knowledge of the amount of inventory,
delivery requirements, and delivery priority scale to ensure the accuracy of the distribution of
each coal quality. The last process in the coal company supply chain is the sales process. This
process could be divided into export and domestic sales. Sales volumes can be used as a
reference for the company to carry out future production, blending, and transportation
planning. As a result, knowledge about the exact number of demands is critical to the coal
industry’s long-term viability, effectiveness, and supply chain connections.
Several studies have been done to improve the effectiveness of the coal industry's
supply chain process. In their research, Belov et al. (2020) focused on the logistic point of
view. They try to improve the transportation process effectiveness in the coal industry's
supply chain by maximizing the transportation plan given the capacity and arrival time of
coal hauling vessels at the port. However, the research has not considered the diverse coal
quality from production, blending, and sales processes.
Peng et al. (2009) did a study to determine coal supply chain optimization using a
combination of washed coal and metallurgical coal. Although this study considers the
4
|
Colorado School of Mines
|
blending process, they look at the raw coal as a bulk rather than dividing them into several
different qualities.
On the other hand, Bauer (1982) focused his research on the coal blending model
without looking at transportation and sales. He is trying to formulate a blending process from
different qualities. However, the study finds difficulties determining a blending formulation
for more than four other coal qualities.
Liu et al. (2000) also discussed the coal blending problem. Their study tries to create
good distribution and allocation decisions while considering each shipment's supply, quality,
and price. But they are looking at the blending process from the customer's point of view.
Thus, the calculation is different from the producer's point of view.
To summarize, all the above research are focusing on different type of supply chain and
not looking at the whole supply chain process in the coal company. An end-to-end model of
coal supply chain process which consider the production, sales, transport, and blending has
not yet been developed.
1.1.2 System Dynamics Modeling
A system dynamic model is a computer-aided method to make better decisions when
dealing with complex and dynamic systems. The goal of the model is to ensure that how the
system works is understood. The feedback system underpins system dynamics, which
compliments the system thinking method. System dynamics, first introduced by Jay W.
Forester, are today used in various system areas, including social, healthcare, biological, and
engineering studies.
The mid-1950s were significantly responsible for Forrester's insights into the
fundamental foundations that underpin engineering and management, which contributed to
system dynamics. Using hand simulations (or computations) of the stock-flow-feedback
structure, which included the actual corporate decision-making framework for hiring and
layoffs, Forrester was able to show how the instability in a company's employment is
attributable to the firm's internal structure. These hand simulations were the beginning of the
study of system dynamics.
Forrester and a group of graduate students sped the move from hand-simulation to
formal computer modeling in the expanding field of system dynamics in the late 1950s and
early 1960s. Bennett (1958) created SIMPLE (Simulation of Industrial Management
Problems with Many Equations), the first system dynamics computer modeling language.
5
|
Colorado School of Mines
|
Continuous with Fox and Pugh (1959) which created the first version of DYNAMO
(DYNAmic MOdels), an updated version of SIMPLE and the system dynamics language that
has been the industry standard for over thirty years. Forrester then published Industrial
Dynamics in 1961, and it is still regarded as a standard in the system dynamics field.
Various system dynamics model has been implemented in a study about the coal
industry. Since the coal industry has a strong relationship with energy generation and the
country's economic condition, many system dynamics studies are trying to model the coal
industry from the bigger point of view, relating them with those factors. The company's
supply chain process has not become a priority on their model. Yet, several system dynamics
studies have also been used to model the multi-product supply chain process similar to the
coal industry. In this study, a combination of coal company's supply chain conditions is tried
to be modeled to support management decisions in giving information about their company's
supply chain condition to enhance their business process.
1.1.2.1 System Dynamics Modeling in Coal Industry
Song et al. (2015) writes research trying to put coal production during a year but not
classify their quality. The model in this study is trying to replicate the relationship between
coal, power supply, and China's economic growth.
Wang et al. (2016) tried to develop those relationships by creating a system dynamic
for coal production in China based on three scenarios: the business-as-usual scenario, the
policy regulation scenario, and the strong policy scenario. In the research, they are modeling
a summary of coal in china without any details in the quality.
There is also research created by Baskoro et al. (2021), aiming to predict the future of
coal production in Indonesia. The model in their study also uses similar scenarios to Wang et
al. (2016) and still does not consider different coal quality.
All of the above research discusses system dynamics in the coal industry but with the
larger viewpoints. Those studies show that coal industry is an exciting topic to discuss using
system dynamics. Although the discussion about detailed coal company’s supply chain has
not been done yet, the discussion relating coal is promising.
1.1.2.2 System Dynamics Modeling for Multi-Product Supply Chain Process
Firstly, Forester (1961) released a book in which one of the chapters illustrates the
production and distribution system dynamics model. He was building a model which includes
the production, inventory, and sales factor inside. The model soon become the supply chain's
6
|
Colorado School of Mines
|
general system dynamics model. However, the illustration was not produced using any
software and was drawn manually.
Sterman (2000) modified the Forester’s idea of the production and distribution model
into a modeling software and discussed it in his book using several different industries study
cases. Some modifications are being introduced to the original model to accommodate the
variety of needs in each industry.
Much research in system dynamics tries to create their industry supply chain based on
the above model. Angerhofer and Angelides (2000) study mentioned the importance of
inventory management using the system dynamics. they found that forester supply chain
model can be improved in those area.
Some of the research also discusses how the inventories could be inputted from a multi-
product source and create a similar case with the coal blending process. Poles and Cheong's
(2009) studies show inventory control in a closed-loop supply chain. In their studies, they use
two products to feed one inventory before their product goes to the customer.
Langroodi and Amiri (2016) also discuss the modeling approach for a multi-level,
multi-product, multi-region supply chain under demand uncertainty. The design for a multi-
area supply chain by drawing on a systemic dynamics approach. They are trying to have a
system that can minimize costs through effective product placement.
The studies above show that system dynamics have a long supply chain management
history. The appearance of system dynamics is driven by anxiety about the company's
ineffectiveness of supply chain management. Many of the models made also show the ability
of system dynamics to handle problems regarding multi-products in companies. Therefore,
implementing system dynamics is considered very promising to explain the supply chain in
coal companies, especially in inventory management.
1.2 Motivation and Goal
The relationship between supply chain process in the coal industry has not become a
focus of much research. With the uncertainty of coal supply, the integrated system is crucial
to create a synergy between each system in the supply chain, especially to ensure precise
product distribution and effectively manage the available stock. Mine planning, production,
blending, transport, and sales are distinct systems in the coal industry's supply chain. Detailed
consideration of each system and the relationship between them should be modeled and
analyzed to provide information about the company's condition. It is also essential to look at
the whole system's behavior to support the future company's decision-making.
7
|
Colorado School of Mines
|
Based on the above facts, an engineering approach is needed to provide a detailed view
of the actual condition, show existing behavior, then analyze future conditions. System
dynamics modeling is considered suitable to meet this need. Until today, the use of system
dynamics in determining the coal industry supply chain has not been reported.
This research aims to create a decision support tool with system dynamics model to
create a model for enhancing the effectiveness of the coal company's supply chain process.
For this purpose, coal companies with varying coal quality reserves and sales are chosen. The
system dynamics model is developed using the available production, blending, transport, and
sales data. Then, the model is validated to ensure its similarity with the actual condition.
Following that, three scenarios are introduced to the model to look at the model behavior in
possible future situations.
1.3 Research Questions
The following are the thesis' research area of interest, as well as the reasons for looking
into possible solutions:
1. Can system dynamics model re-create the coal industry supply chain with a multi-
product blending process? And can the model give the information about the
remaining product in each process?
To answer this question, a system dynamics model is created with the coal
company’s supply chain available data. After that, an analysis of the model made is
carried out to ensure that the model can run without any error. Then, the result is
observed to know whether it is showing the remaining inventory in the system
2. What is the level of accuracy between the original model and the real condition?
To answer this question, all steps leading up to the simulation step in system
dynamics will be completed. After that, we’ll compare the model’s behavior to the
actual situation’s behavior and calculate the accuracy. From these results, we can
see how much confidence we have in the model.
3. What are the scenarios that are introduced to the model, and what are the results of
each scenario executed?
The next step is to develop a scenario based on the possibilities that arise in the
model, which is a follow-up to the model already in line with the actual situation.
Three different scenarios were provided to the model to establish its behavior in this
investigation.
8
|
Colorado School of Mines
|
4. What is the relationship that can be seen from the results obtained?
Following the introduction of the scenario and knowledge of the model’s behavior,
the final step is to analyze what interpretation can be concluded from all the results
that have been obtained.
1.4 Original Contribution
Coal company’s supply chain detailed model has not yet been made using system
dynamics approach. However, many business challenges are solved through the application
of system dynamics. The term “system dynamics” is commonly used to describe the
resiliency of a system to changes in the business environment. Several scholars have also
used system dynamics to evaluate inventory within their own companies. This research
provides the coal company’s supply chain model by looking into detailed data from
production, sales, transport, and blending.
It is known that system dynamics model requires to have relationship among the
component. Some researchers attempted to develop their own equation in their model’s
component within the system dynamics model. However, the coal industry has its own
uniqueness in their supply chain process. Coal has a quality range with different designations,
and every powerplant or factory that requires coal as its energy source has diverse coal
quality specifications. It is also possible to have a blending process that companies can do to
fulfill these different needs in this coal industry. Therefore, in this thesis a formulation and
relationship between components are created to determine how a company can manage its
coal reserves to meet existing market needs optimally.
In the system dynamics studies, usually there are some scenarios being introduced to
the model to ensure the model works with future predicted conditions. This thesis also
provides three scenarios namely diversification, 2022 supply-demand, and 2023 supply-
demand, based on the company’s long-term business plan.
1.5 Thesis Structure
This thesis consists of nine different chapters. The first chapter is an introduction that
explaining the problem, state of the art, motivation of the research, research question, original
contribution to the topics, and the thesis structure. The second chapter describes a
fundamental of system dynamics, which cover theoretical information about the method. The
third chapter is discussed about the description of case study that covers how the research
9
|
Colorado School of Mines
|
CHAPTER 2
FUNDAMENTALS OF SYSTEMS DYNAMIC MODELING
System dynamics is a tool that helps understanding of complex systems behavior over
time (Sterman, 2000). He mentioned that system dynamics is based on the theory of
nonlinear dynamics and feedback control created in mathematics, physics, and engineering
because as the focus is modeling the behavior of complex systems.
The goals of system dynamics, according to Sapiri et al. (2017), are to improve
decision-making processes and policy information by including all important cause-and-
effect relationships, time delays and feedback loops in dynamic behaviors. They also
mentioned that if a system dynamic is done correctly, we may establish a loop of interaction
between parameters, or what we call variable and stock in system dynamics, to better
understand the structure and behavior.
Sapiri et al. (2017), also note that system dynamics have various properties that
distinguish them from other methods, including:
1. Having a holistic perspective to assist in strategic decision-making;
2. Capable of assembling and homogenizing entities;
3. Model the system using a technique which allow one to see interaction between
various components in the system;
4. Able to simulate the behavioral change between each components;
5. Finding a root cause to a problem from behavior pattern; and
6. Can convert a qualitative variable to a quantitative variable.
Building a system dynamics model, requires understanding the concept of the system
and its complexities, which helps in modeling an existing system explicitly.
The creation of a system is the most essential part in system dynamics. Any group of
pieces that interact, are interconnected or are interdependent to produce a complex and
cohesive whole with a specified purpose is referred to as a system (Kim, 1999). According to
Kaufman (1980), a system functions when its behavior is determined by its complete
structure rather than the sum of its individual components. To complete a task, Kaufman
adds, all components must be properly placed, as shifting just one or a few will cause the
system to fail.
Ladyman et al. (2020) defines complex systems by distinguishing the factors that
differentiate complexity. The number of interactions between components is a primary
11
|
Colorado School of Mines
|
characteristic of a complex system; moreover, the complex system is not coordinated or
controlled centrally, and any component may differ from the others.
It is also known the interaction of a complex system that has feedback within the
timescale iterated. Complex systems are exposed to the outside world and are controlled by
external factors. Ladyman et al. (2020) also pointed out that the complex system is
nonlinearly dependent but remains stable in the face of relevant disturbance. Aside from that,
the system has various scales of functioning structure, can store historical data, and has
adaptive behavior following their surroundings.
Kaufmann (1980) defined a complex system as one that exhibits 10 distinct behaviors
when compared to a simple system. Those are as follows:
1. Self-stabilizing – complex systems can maintain self-stability in the face of
changing environmental conditions.
2. Aim-seeking —complex systems have a specific goal that it will actively pursue.
3. Program-following — complex systems always follow their program.
4. Self-reprogramming — As the system becomes more complicated, it adapts to the
new environment and avoids errors.
5. Anticipation – complex systems predict mechanism of systems` change.
6. Environment modification – complex systems try to simplify the environment to
make it easier to run.
7. Self-replicating –complex systems have ability to reproduce or copy itself.
8. Self-maintaining and repairing – complex systems continually change and
maintaining itself.
9. Self-recognizing — When complex systems recognize themselves, they can
restructure and adjust their connections.
10. Self-programming —complex systems can establish their own goals and solve
them in their way.
However, not all complex systems have the characteristics mentioned above. For
example, number 7 and number 8 are frequently found in the life system. The complex
system could not be fully comprehended by focusing solely on a single parameter or problem.
There is feedback to feed the system as each parameter controls the others, which may aid the
system in making its forecast. Developing mathematical models of complex dynamic systems
needs more than just technical tools (Sterman, 2000). To model a complicated system, system
thinking is needed to analyze the components thoroughly.
12
|
Colorado School of Mines
|
System thinking is a technique that allows one to visualize thoughts using a tool.
System behaviors can be better comprehended and predicted by using strategies of reasoning
and then implementing them (Waters Center, 2021). Causal Loop Diagrams (CLD), Stocks
and Flows and Feedback Loops are tools used to describe a complex system in system
thinking.
The system should include a separation between the parameters to represent what has
been included in the model. The term “stock and flow” is used in system thinking to
distinguish between the collected parameters and the flow parameters. Dynamic thinking is a
method of removing non-essential elements from a model, and it enables the issue to be
described in a behavior pattern throughout time. Operational thinking skills are required to
construct a stock and flow diagram effectively, and the ability to represent differences among
parameters will be developed through operational thinking.
Finally, to think about systems, we must depict the relationship between the elements.
A ’problem’s cause-and-effect relationship should build a loop and a closed-loop system. A
skill that could reflect this relationship is closed loop thinking. The system is reinforced by
the nonlinear thinking skill, which acts as a third variable. To mimic the behavior of a
complex system, the system dynamics model is used to visualize the system thinking
approach.
2.1 Steps in System Dynamics Modeling
According to the Sterman (2000), numerous critical phases must be considered while
developing a system dynamics model, including describing the problem, articulating the
hypothesis, creating a dynamic model, designing the scenario, validating the model, and
evaluating the scenario. In the first phase, we look at the ‘problem's core issue, the ‘model's
potential audience, and the ‘model's aim. After determining the fundamental problem,
identifying relevant parameters within the model itself is needed. The modeler must capture
all processes that may affect the model's behavior later and develop a basic framework for the
model in this step.
The establishment of a system dynamic model involves all previously specified
parameters. Then, the scenarios which can happen in the future are designed. This designing
scenario step includes determining their component and equation for later use. The next stage
is to simulate the model. The first step in simulation is deciding whether the model is
accurate and true to its original state. It can be tweaked up to the level of most resemblance to
13
|
Colorado School of Mines
|
the actual scenario. The next stage is to test the scenario introduced before and see how the
model's behavior would vary based on the options.
The first step in creating system dynamics model is understanding the challenges that
need to be solved. It is hoped all the problem’s components to be discovered by that point.
This identification is essential for determining the source of the problem and the variables
used in the model (Sapiri, 2017). According to Sterman (2000), identifying the type of the
variables is equally significant when determining the variables.
In addition, determining the reference in model creation is critical in this initial phase.
The reference can be used as a base line for investigating the model’s behavior in the future.
If there is a behavior disparity at the start of the model, it should be identified and
determined. This, too, is intended to ensure the model behaves in conformity with reality.
Furthermore, selecting the time horizon is a critical component of the modeling process’ first
stage. The model’s needs are considered when determining the time component, so the results
are described subsequently according to the appropriate period.
The final stage is to figure out how big the model should be. The key model’s scope is
set here to keep the model focused and in line with its aims.
The next step in building a system dynamics model is to develop an initial hypothesis
and link it to all previously identified parameters. A relationship is established to introduce
the system’s feedback structure; the causal loop diagram (CLD), is the instrument for
introducing this feedback. The interaction between components and how each component
affects other components may be seen in this process, and the initial framework for the final
model can be displayed at this stage.
Constructing stock and flow diagrams is the next step in creating the system dynamics
model. The CLD is extended with stock and flow diagrams. Components can accumulate
with a delay function when stock and flow diagrams are created. This stage also includes
entering all the model’s equations, ensuring it is ready for simulation. Causal Loop Diagram
is the model’s final diagram, implying the simulation step is the next stage. This stage can
uncover possible errors in the previous steps and ensures all the previously completed
concepts can be integrated into a computer program (Sterman, 2000).
The scenario design process is done by creating a situation which could be introduced
to the model. The process is made by looking into possible condition that could happen to the
model. Several new components can be later introduced to the model depending on the
definition of the scenario. The scenarios are used as the validation to determine the model
ability to take new condition without any errors in the structure.
14
|
Colorado School of Mines
|
Model validation is the process of simulating all the models that have been constructed
to ensure that they work smoothly. There are two crucial aspects to consider in this step:
structural validation and behavior validation (Sapiri, 2017). This validation structure ensures
that all model components and interactions agree with the problem at hand. On the other
hand, behavior validation guarantees that the model's equation is correct. Simulation helps to
examine how the model behaves. In the model validation process, there are two steps that
need to be conducted, the accuracy assessment and the scenario testing.
In the accuracy assessment there are three different methods to determine the model's
accuracy. The accuracies are calculated based on the model's performance in issuing output
(predicted value) that is observed in actual (real) behavior of the system using metrics. The
first metric is absolute accuracy that can be seen in Eq.2.1:
4(,%
(1+(2 < 4(,%),78 9 :;100%?
1+(2
!"#$%&'( !**&+,*- = / (2.1)
1+(2
(1+(2 > 4(,%),78 9 :;100%?
4(,%
where:
Pred = Predicted value in the system
Real = Real value in the system
The calculation of absolute accuracy is made based on comparing the output value with
the actual value. If the output value of the system is less than the actual value, the real value
is divided by the prediction value and vice versa.
The second accuracy metric uses the Root mean squared error (RMSE) method. RMSE
describes the extent to which the predictions made by the system have errors. RMSE can be
calculated using Eq.1.2:
(1+(2 – 4(,%)!
RMSE = EF G J (2.2)
I
where:
RMSE = Root Mean Squared Error
Pred = Prediction Value
Real = Real Value
S = Sum
n = Sample size
Eq.1.2 suggests that the RMSE can be interpreted as a normalized distance between the
vector of expected values and observed values. It allows determination of the standard
deviation for a typical single observation instead of some total error. The RMSE is helps to
15
|
Colorado School of Mines
|
determine how far off the model is on its next forecast. The smaller the RMSE obtained, the
better the model predicts the actual situation.
The third accuracy metric is mean absolute percentage error (MAPE). MAPE shows
how much error is generated from the predicted value compared to the actual value. The
equation used in the MAPE calculation is given in Eq.2.3:
"
1 4(,% −1+(2
MAPE = ;100%GM M (2.3)
I 4(,%
#$%
where:
MAPE = Mean Absolute Percentage Error
Pred = Prediction Value
Real = Real Value
S = Sum
n = Sample size
The smaller the MAPE, the more accurate the prediction made by the system. With all
these accuracy metrics, a set of forecasts from the system is generated for each component
which is then ranked using all accuracy metrics. This helps in determining which component
has the highest level of confidence from all existing ranking results.
The scenario evaluation process involves feeding various new situations to the model
to observe how it reacts and how sensitive it is to a new problem that may arise. With the
scenario evaluation, the result could be compared and shown how each scenario affects and
benefits the model. Later, the analysis and conclusion could be determined made based on the
scenario evaluation process.
2.2 Diagrams in System Dynamics Modeling
System dynamics uses two diagramming tools to visualize a system. One is the causal
loop diagram (CLD), and the other is the stock and flow diagram (SFD). Both diagraming
tools are essential in modeling system dynamics. They both have different goals and
functions when presenting the system. In this section, we are going to look at both CLD and
SFD.
2.2.1 Causal Loop Diagram
A causal loop diagram in system dynamics is the simplest method to visualize the basic
ideas of a system. The causal loop diagram can be used as the first model to determine
parameters and the relationship between them. The causal loop diagram is also a great
16
|
Colorado School of Mines
|
diagram to be shown to the end user or people with little or no understanding to the system
because of its simplicity.
From the Figure 2.1 we know that there could be a positive and negative feedback
resulting from the components. The causal loop diagram is the ideal tool for capturing
hypotheses, eliciting, and capturing the mental model and communicating the critical
feedback that causes a problem (Sterman, 2000). He added, however, that the diagram has
some limits. The first limitation mentioned by him is that a causal loop diagram can never be
comprehensive. Because the causal loop is used to simplify the model, this is the case. The
causal loop diagram is never final, but always remains provisional. The diagram can always
be modified, improved, and evolved based on the issue. However, he also mentioned that
CLD does not discriminate between stock and flow, and because of this, it cannot be turned
into a more detailed model.
Figure 2.1 Basic Causal Loop Diagram (Sapiri et al, 2017)
Sapiri et al. (2017) explains to verify that the dynamics relationship exists in the CLD,
at least one of the relationship chains must be a closed loop. When creating an SD model, the
loop needs to be able to help us describe the system's structure. CLD represents cause and
effect relationships between system variables and, when connected, forms closed loops.
2.2.1.1 Feedback
The feedback concept is at the heart of the system dynamics approach (Dangerfield,
2020). Dangerfield indicated that a feedback loop exists when information resulting from
some action travels through a system and eventually returns in some form to its point of
origin, potentially influencing future action. Most of system dynamics modeling is aimed to
discover and represent the feedback process which, along with stock and flow structures, time
delays and nonlinearities, determine the dynamics of a system (Sterman, 2000). Hence
17
|
Colorado School of Mines
|
feedback is the main element of a system dynamics model. The feedback or interactions
within the system are usually what makes the system complex. The feedbacks are two types:
positive feedback and negative feedback.
Positive feedback, or reinforcing feedback, is defined as self-reinforcing feedback.
When one parameter is rising, the other will be rise as an effect of the first, and vice-versa. In
a case of the chicken and egg in Figure 2.2a, more chickens will lay more eggs, and more
eggs also lead to more chicken.
Figure 2.2 Feedback process with (a) positive loop and (b) negative loop (Sterman, 2000)
Conversely, negative feedback is considered self-correcting, which will act as a balance
in the system, counteracting the changes in the system. In a case of a chicken and road
crossing (Figure 2.2b), it is illustrated that if there are more chicken, the probability of
chicken crossing the road is higher, and chickens dead from traffic leads to fewer chicken. It
can then be said the road crossing is having a negative effect on the chickens and eventually
will drop the number of chickens to zero. Sterman noted that all systems, no matter how
complex, consist of networks of positive and negative feedback, and all dynamics arise from
the interaction of these loops with one another.
Feedback Loop
There may be a lot of feedback in the system dynamics modeling, as the system has its
own reinforcing and self-correcting parameters. When this happens, the system creates a
complex pattern or dynamic, and the interaction between parameters and results in unique
behavior. When a system creates a closed path, it is called a feedback loop. Forrester (1968)
18
|
Colorado School of Mines
|
describes a feedback loop as a closed path connecting in sequence in a decision that controls
action, system level and information about the system level that will later return it to the
decision-making point, as illustrated in Figure 2.3.
Figure 2.3 Feedback Loop (adapted from Forrester, 1968)
Forester (1968) said of the figure that, with the sources regulating the current decision,
the decision will cause an action, which will then modify the system’s level. Next, the system
will generate information as its next stage. He emphasized that the information could be late
or contain errors, describing its apparent level as opposed to its genuine level. However,
knowledge remains the foundation for the decision-making process (Forrester, 1968).
2.2.2 Stock and Flow Diagram
The stock and flow can be used to input the stock and flow functions into the model,
with its flow function providing the accumulation of parameters made into incoming and
outgoing stock. Additional components in the causal loop diagram, such as connectors and
auxiliary variables, serve as a supplement to the stock and flow model’s function. Stock and
flow diagrams are the most utilized diagrams in system dynamic models. They provide a
visual understanding of the system’s behavior. Stock and flow diagrams also allow the model
to be exposed to various scenarios to see how the system changes.
Compared to CLD, stock and flow diagrams have various components, as previously
stated: Stock, flow, auxiliary variables, and connectors. Table 2.1 explain the stock and flow
component which adapted from Sapiri et al., (2017) and Sterman (2000).
19
|
Colorado School of Mines
|
Table 2.1 Stock and Flow Diagram Components (modified from Sapiri et al., 2017)
Elements Description Symbol
Stock Stock is a model element built up from each time
step. it is typically a component with a quantity, such
as inventory.
Flow Flow is a component that works as input and
output for the stock element's value. it is the top link
for the stock element. Clouds at the beginning and
end of the flow represent the outside boundary of the
model
Auxiliary Auxiliary is a sub-element used to combine stock
elements into more minor elements. It is an element
that does not accumulate on a timestep basis.
Connector The connector element provides links of cause and
effect to the auxiliary or the stock elements.
A simple stock and flow diagram of an inventory is shown in Figure 2.4, where an
inventory becomes stock in the model. There are also flow elements, namely inflow and
outflow, and the desired rate is connected using a connector to provide a value for inflow that
depends on outflow and inventory.
Figure 2.4 Simple stock and flow diagram (modified from Sterman, 2000)
According to Sterman (2000), who cites Mass (1980), stock and flow diagrams are
particularly important in the system dynamics model because they may show the dynamics of
a system, which include:
1. Laying the groundwork for activity;
2. Providing the system with inertia and memory;
3. Providing a source of delay; and
20
|
Colorado School of Mines
|
CHAPTER 3
DESCRIPTION OF CASE STUDY
This study includes a complex multi-products coal industry’s supply chain created with
system dynamics model. Object of this study is taken from one of Indonesian coal company
with variety of mine brand and market brand. The study will be using their production,
transport, blending, and sales data as the main data. The mining method used in this company
is surface strip mining which is done by first removing a long strip of overlying soil and rock
and then extracted the resources. This method allows all the results of coal mining to be
transported using trucks from the mining site to the stockpile location. The total production of
the company is 35.9 Mt. In this study, sales component is divided into two types of deals:
export and domestic. In both, the sales method used is FOB (free on board), allowing the
company to only be responsible for the product until it enters the buyer’s ship. The total sales
of the company are 29.3 Mt. Besides sales, the company also using the coal for their own
needs with total of 1.2 Mt.
3.1 Supply Chain in the Company
The company have complex supply chain process which consist of planning,
production, inventory, blending, and sales. Illustrated in the Figure 3.1 below is the general
process of the coal company’s supply chain process.
Planning Production Mine Inventory Train Transport Port Inventory Sales
Plan a strategy Extracting Managing Moving materials Stockpiling, Selling materials to
according to the MaterialFsigure 3In.v1e nCtoroya lelv Celso mpantyo' sth eS puoprtply Chaliona ding and the customers
long-term plan and and quality managing sales
sales blending
The supply chain process is started by planning the production based on the long-term
plan and sales. After the planning, the process continues to the extraction process. The
process then followed with transport to the mining inventory. In the mining inventory,
blending process is conducted using 11 mine product combinations that can blend into five
different sales products. After the blending, the process continues with transporting the coal
to the port, which is done via train. The mine stockpile transports the coal according to the
blending specification requirements, adjusting the quality with the desired market brand.
22
|
Colorado School of Mines
|
After arriving at the port, all coal is placed according to the market brand determined on each
port stockpile and then transferred to the buyer’s ship.
In practice, there is some problems that occurred in this company’s supply chain
process. The first problem is the blending planning process is still done by distributing the
mine product manually using Excel, considering the sales demand. There is often a difference
in the volume recorded at the time of registering at each different supply chain process.
Second, there is also a time when the production or transport is delayed to a certain condition.
The delayed process will disrupt the continuity of the supply chain process. There is also one
term in the sales process, referred to as a spot sale. In spot sale, the buyer has a sudden
demand to the companies without any long-term planning before.
Those sudden movement in the production or sales value are very difficult for the
company to predict and plan. This ultimately leads to adjustments and other inspection
processes that may disrupt and drag the business decision. Therefore, it is essential to have an
exact and accurate knowledge of the remaining coal specification to create a fast decision
regarding above problems. Thus, an integrated system which can show real-time conditions
and predictions, is beneficial to help the company in making a faster and efficient decisions.
3.2 Scenario for the Model
After the model has been established, scenarios are introduced to the model. In this
study, Scenarios are used to predict the company's future business planning. Therefore, the
company's long-term planning data are utilized. The scenarios are Diversification Scenario,
the 2022 Supply Demand Scenario, and the 2023 Supply Demand Scenario.
The first scenario (business diversification) created based on the company's plan to find
more environmentally friendly alternatives from coal. The distribution will have more
components to consider with this energy transition plan, and therefore, modification of the
model is needed. The first scenario is created to look at the company's ability to handle this
new business expansion demand.
The second and third scenario is created to look at the company's plan stability in
planning their supply and demand. With the new supply and demand rate, the system's
behavior is analyzed and looked at if there is any difference from the original model supply
and demand.
23
|
Colorado School of Mines
|
CHAPTER 4
RESEARCH METHODOLOGY
To analyze production and sales scenarios, this research develops the system dynamics
model using Vensim PLE+ software. The modeling first validates the real data to ensure
similarity, and later several scenarios are introduced to the model for testing purposes. The
flow chart outlines the step-by-step work for the research and is provided in Figure 4.1.
Figure 4.1 Flowchart of the research
The first step in this research is to do a literature review to determine the best strategy
for solving the problem. After determining the approach, data is collected, in this case
utilizing available company data. The following stage is pre-processing, which involves
converting the available data into the same format. The allocation and placement of each
piece of information are also determined during the pre-processing processes. Following the
pre-processing step, the modeling and scenario-defining steps will be completed in
simultaneously. After that, the model is validated to check how accurate it is in comparison to
the actual situation. The final stage is to Simulation scenarios in order to examine the
findings and draw conclusions about the current situation.
25
|
Colorado School of Mines
|
4.1 Data Gathering
The data gathering process is done by getting the data from several different working
units within an Indonesian coal firm. The model’s production input was its planning and
implementation of coal production in 2021.The sales figures originate from the actualization
and forecasting of sales for that same year. The two forms of sales statistics provided are
domestic and export sales. Other than production and sales data, there is also inventory data,
taken from the last inventory in December 2020 and then introduced to the model’s initial
values. For data validation purposes, data from company’s transport realization is also being
used.
Three scenarios are introduced to the model later, the first being a coal diversification
strategy that is part of its long-term company outlook. New demand and their required
calculation (desired rate, recovery rate, and initial inventory) for a coal diversification plant
data are taken for this first scenario purposes. In the second and third scenarios data from the
company’s long-term goals for 2022 and 2023 are used.
4.2 Pre-Processing
The next step is pre-processing the obtained data from the company. Production data is
compiled into 11 types of mine products ranging from the lowest quality to the highest, and
production is then divided into monthly data based on the realization of the actual data. The
market brand is taken from the sales data and divided into five different market products. The
range of the market brand follows the global index quality and buyer requests.
Within the inventory data are two types, one for the mine’s inventory, which is stored
in its product inventory, and the port inventory, which stored at the port. The data is
subsequently calculated with the same undisclosed parameters to retain the confidentiality of
the original data. It is then combined into one file to create input for a later model.
After the datasets are ready, the next step in pre-processing is to define the parameters
to be used, a determination which is carried out with the goal of ensuring all problems are
mapped properly before entering the modeling process. At this stage, the possible
relationship of each parameter is also important to consider in order to simplify the modeling
process.
26
|
Colorado School of Mines
|
4.3 Modeling
The modeling process in this research uses Vensim PLE+ software. The first step is
creating new parameters and putting links between them; the parameters are then given the
equation in accordance with the real situation. The modeling process also creates a timestep
to be used in the model. In this research, the timestep is monthly, and the total time is one
year.
There will be simplistic model shown in the beginning of the chapter to show how the
system generally works and links. The model used the causal loop diagram, which
summarizes the problem against the actual model. The causal loop diagram does not contain
all flows or auxiliary variables.
Later, the causal loop diagram will be translated using the Vensim PLE+ into a stock
and flow diagram, which is the completed model that considers all parameters and can be run
to simulate the problem and scenario. In the stock and flow diagram, the model will become
more complicated since there will be more flows and auxiliary variables. The process of
translating a causal loop diagram to a stock and flow diagram involves identifying the levels
of the system and adding the necessary flows (some of them are included in the causal loop
diagram and some are not); all other variables are auxiliaries.
4.4 Defining Scenario
Scenario determination is used to produce scenarios that will be used in the main
model. This is done to foresee future possibilities, allowing the defining scenario process to
be carried out parallel to the modeling step. After the scenario is determined, it is linked to
the model structure already in place. Several scenario components are created at this point,
and the linkages between components and the existing model are also determined. The
equation of the scenario’s components is also being created by one by one. The produced
scenario is then integrated into the model framework and the existing model’s equations are
adjusted.
4.5 Model Validation
After the model has a complete structure according to the components, relationships
and equations that have been previously planned, the model validation stage is performed.
The previously constructed model attempts to execute in this initial model testing stage.
This simulation is carried out to check that the model’s behavior matches that of its
real condition.
27
|
Colorado School of Mines
|
Accuracy calculations are also done in this step; they give us information how the
model works and how high the confidence level is for the model. In the modeling validation
step, we also need to ensure the proper equation appear in the model and are applicable not
just for the current context, but also for other conditions. This is done to see if the model can
accommodate a variety of situations and the scenario that will be introduced.
4.5.1 Accuracy Assessment
The accuracy calculation is carried out by comparing the model result with the actual
data. The accuracy calculation is done to the input and output component, transport
component, and the mine stockpile components. Analysis of these components is completed
to ensure that every component is correct and has connected behavior. The Absolute
Accuracy, RSME, and MAPE method are used for the calculation. After the analysis, the
result is then sorted into ranks. All ranks are then being analyzed to see if there is any
noticeable pattern and provide important information regarding the model.
4.5.2 Scenario Evaluation
Scenario evaluation is used to observe how the model’s behavior varies when different
scenarios are presented. At this point, a test is carried out on the previously created scenarios;
conclusions and an analysis of how robust the model is in the particular scenario can be taken
from the outcomes of these tests. The scenarios are initially carried out one by one in
this stage to confirm all the components’ equation work as expected and that an analysis can
be performed together with the initial model. After all scenarios have been proven, the next
step is to run simulations with all scenarios together to ensure the model can generate
behavior when all scenarios are engaged at the same time.
4.6 Results
The model and all possibilities gained in the simulation process can be analyzed at this
point. This computation seeks to provide a comprehensive description of the previous stages’
results, as well as explanations and observations of the model’s behavior. It is also possible to
determine the model’s flaws and the corrective actions that must be made for the future.
The most essential thing at this point is that conclusions may be formed from the
conducted simulation, as well as advice to the company if there are any discrepancies in the
obtained results. If coal volumes are insufficient after the scenario is carried out, for example,
the procedures that must be prepared and executed to anticipate this situation are specified.
28
|
Colorado School of Mines
|
the model are then made. The basis for this comes from the supply chain structure of the coal
company being used in this study and company long-term agenda.
5.2.1 Components of the Model
In making the system dynamics model, several components of system dynamics are
introduced according to their designation to create a system that can run according to the
actual situation. Components of the model are based on the company’s different supply chain
sectors that are used as a study case in this study. The model is divided into four main
components:
1. Input/Output components, the components which have equation that contain input
and output data from the company;
2. Inventory components, components containing the results of calculations from input
and output data that have been processed;
3. Connector components, core components of the system because they contain all the
main equations that can connect the entire system to resemble the behavior of its
original state; and
4. Variable components, which help the connector components create supplementary
equation to satisfy the main equation of the model.
Each component has a different equation and function. Each component will be
explained along with the equations in the sub-chapters below to better understand the model.
5.2.1.1 Input / Output Component
Input and output components are made to generate data from company databases into
the software. This component is made to allow for data updates and extend or change the
simulation time. The Input and Output components in this study are divided into three main
components, each of components will be further divided into sub-components according to
the quality of the existing coal. The three components are the production component, and the
sales component, which will be further feed from exports and imports, and the internal usage
component. These three components are the end of the system where the entire simulation
will start and end from the data that comes from this component.
Production Components
The coal that has been produced is carried by truck to the stockpile as part of the
production component. Monthly production data for the components of this model is
31
|
Colorado School of Mines
|
Although these sales components are considered as the output components, since the
company have two different sales data, the component is taking the data from the domestic
sales and export sales components rather than using the same equation with the production.
Therefore, instead of adopting the ‘OPQ RST U!Q!’ function, the equation for sales
component is:
T,%(# = PT +UT (5.2)
- - -
Where:
Sales = Sales volume of market brand j (j = 1,…,5)
j
ES = Export sales volume of market brand j
j
DS = Domestic sales volume of market brand j
j
j = 1:LC; 2:LMC; 3:MC; 4:MHC; 5:HC
The equation above explain that the sales component will generate its value based on
the export sales and domestic sales. It is the export and domestic components that will use the
excel retrieval function later. The Eq.5.2 then translated into each market brand resulting
detailed equation in the Table 5.2.
Table 5.2 Sales Component Equation
Market Brand Vensim PLE+ Equation
LC LC sales = (LC Dom + LC Ex)
LMC LMC sales = (LMC Dom + LMC Ex)
MC MC sales = (MC Dom + MC Ex)
MHC MHC sales = (MHC Dom + MHC Ex)
HC HC sales = (HC Dom + HC Ex)
From the equation above, the sales components are the result of the sum of the export
and domestic sales components of each market brand. The domestic and export sales are later
explained in the variable components in chapter 5.2.4.1.
Internal Usage Components
There is also an internal usage component which contains the coal used for the
company's internal needs. The company's internal usage components also use equations to
retrieve data from an excel database similar to the production component. This component
only has 1 component that contains the equation showed in the Table 5.3 below:
34
|
Colorado School of Mines
|
As previously stated, no database is used in this component; rather, this component is
the result of computations using the following fundamental Eq.5.3 below:
%%
ZT = 1+$2 −[GQ+,WI \ (5.3)
& & &
&$%
Where:
MS = Mine stockpile volume of mine product i (i = 1,…,11)
i
Prod = Production volume of mine product i
i
Σ = sum
Train = Train transport volume from mine product i
i
Variety of transport component in each mine product resulting different equation for
each mine products. The complete equation for each product can be seen from the Table 5.4
below.
Table 5.4 Mine stockpile equation for all mine products
Mine Product Name Vensim PLE+ Equation
MP1 MP1 MS = (MP1 Prod) - MP1 for IU Stock - MP1 Train 1 - MP1
Train 2
MP2 MP2 MS = (MP2 Prod) - MP2 Train 1 - MP2 Train 1
MP3 MP3 MS = (MP3 Prod) - MP3 Train 1 - MP3 Train 1
MP4 MP4 MS = (MP4 Prod) - MP4 Train 1 - MP4 Train 2
MP5 MP5 MS = (MP5 Prod) - MP5 Train 1
MP6 MP6 MS = (MP6 Prod) - MP6 Train 1 - MP6 Train 2 - MP6 Train 3
MP7 MP7 MS = (MP7 Prod) - MP7 Train 1 - MP7 Train 2 - MP7 Train 3
MP8 MP8 MS = (MP8 Prod) - MP8 Train 1 - MP8 Train 2
MP9 MP9 MS = (MP9 Prod) - MP9 Train 1 - MP9 Train 2
MP10 MP10 MS = (MP10 Prod) - MP10 Train 1 - MP10 Train 2
MP11 MP11 MS = (MP 11 Prod) - MP11 Train 1 - MP11 Train 2 - MP11
Train 3
Table 5.4 shows the different equation based on train transport components that exist in
each mine brand. there is mine brand with only one train transport component, to mine brand
with three train transport components. The train transport for each mine brand, will later be
explain in chapter 5.2.1.3. However, in terms of production, there is only one component in
every mine brand.
36
|
Colorado School of Mines
|
Therefore, besides above equation, the mine stockpile components also contain the
initial value of each mine brand. The data for the initial value is taken from the inventory
report data from the previous year. When the simulation is Simulation, this initial value data
will become addition value for the production in TIME 0 and then continue to be computed
using the equation above. Table 5.5 shows the remaining stockpile inventory data in January
2021 based on the company’s data.
Table 5.5 Initial mine stockpile value (tonnes)
Mine Product Mine Product Stockpile Stock
MP 1 152,897
MP 2 40,000
MP 3 0
MP 4 12,000
MP 5 0
MP 6 0
MP 7 0
MP 8 0
MP 9 0
MP 10 139,540
MP 11 80,000
Table 5.5 shows that not all mine products have inventory remaining at the beginning
of 2021; only five qualities have inventory at the beginning of the year. MP1 products have
152,897 tonnes which is the largest inventory held by all mine products. Furthermore, there is
also the MP10, which has an inventory of 139,540 tonnes. Apart from these two qualities, the
three remaining inventories are not more than 100,000 tonnes, namely MP11 with 80,000
tonnes, MP2 with 40,000 tonnes remaining, and MP4 with only 12,000 tonnes remaining All
other qualities have no remaining inventory.
Port Stockpile
Slightly different from mine stockpile components, the port stockpile components are
named after the company’s coal market brands rather than its mine brand. These components
get their value from the total volume transported coal from the mine stockpile reduced by the
coal sales volume. According to the market brand available on the model, this component is
37
|
Colorado School of Mines
|
separated into five different components. The five components of the stockpile port are as
follows:
1. Low Coal (LC) Stockpile
2. Low-Medium Coal (LMC) Stockpile
3. Medium Coal (MC) Stockpile
4. Medium-High Coal (MHC) Stockpile
5. High Coal (HC) Stockpile
As previously stated, the calculations for this component are based on mine transport of
specific coal brands minus by the sales in the coal brand market. As a result, the general
equation for this component can be stated as Eq.5.4 follows:
%%
PS = [GQ+,WI \−T,%(# (5.4)
. &,- -
&$%
Where :
PS = Port stockpile volume of market brand j (j = 1,…,5)
j
Train = Train transport volume to mine brand i (i = 1,…,11)
i
Σ = Sum
Sales = Sales volume of market brand j
j = 1:LC; 2:LMC; 3:MC; 4:MHC; 5:HC
Each of the components in this port stockpile have varied equation depending on the
capability of each mine product to be blended into the market brand. Table 5.6 provides a
more extensive look at the equation.
Table 5.6 Port Stockpile Component Equation
Market Brand Name Vensim PLE+ Equation
LC LC PS = (MP2 Train 1 + MP3 Train 1 + MP4 Train 2 + MP1 Train
1) - LC Sales
LMC LMC PS = (MP1 Train 2 + MP2 Train 2 + MP3 Train 2 + MP4
Train 1 + MP5 Train 1 + MP6 Train 3 + MP7 Train 3) - LMC Sales
MC (MP10 Train 2 + MP11 Train 3 + MP6 Train 1 + MP7 Train 1 +
MP8 Train 2 + MP9 Train 2) - MC Sales
MHC (MP10 Train 1+MP11 Train 2+MP6 Train 2+MP7 Train 2+MP8
Train 1+MP9 Train 1)-MHC Sales
HC MP11 Train 1-HC Sales
38
|
Colorado School of Mines
|
We can see which mine product feeds each market brand from the table. Besides a
value generated by the equation above, the port stockpile also has an initial value inside the
component. This initial value (remaining stock) will be entered the calculation in TIME 0, in
the same way as the mine stockpile. Table 5.7 shows the remaining inventory from the
previous year by port stockpile component.
Table 5.7 Port Stockpile Remaining Inventory
Quality Port Stockpile
LC 7,599.07
LMC 240.46
MC 0
MHC 0
HC 21,593.1
The HC component had the most remaining coal with 21,593.1 tonnes in early 2021,
followed by LC and LMC, according to the remaining inventory data. On the other hand, MC
and HC had no remaining stock in 2020.
Internal Usage stockpile
The internal usage (IU) stock component is the last inventory component. It only
consists of one component like the mine stockpile component. The IU component uses the
equation stated in Table 5.8 below:
Table 5.8 IU Stock Component Equation
Market Brand Name Vensim PLE+ Equation
IU Stock IU Stock = MP1 for IU Stock - Internal Usage
The IU is only using one mine product as its feeder to satisfy the internal needs.
Therefore, there is only one transport component included in the equation. Same as other
inventory components, this initial value also introduced in TIME 0 of this component
calculations.
5.2.1.3 Connector components (Transport)
The connector component is one that serves as a link between other components. The
IF THEN ELSE function is the function used for connector components. The function is used
39
|
Colorado School of Mines
|
because the component requires conditions where there is some prioritization, and different
usage of values in each component.
The train transport component is the connector component under which coal from the
mine stockpile is transported to the port stockpile in this case, via train transportation. In the
making of train transport component, each mine product's priorities and capabilities are
determined. There is a prioritize mine brand for each market brand product. If the prioritize
mine quality coal cannot satisfy the market brand's desired volume, another mine brand
product needs to be introduced. As a result, every potential mine product is linked to a market
brand by the train transport component to anticipates the above condition.
Eventually, the System will try to feed the market brand with the mine product based
on the prioritization we set earlier. So that, if it remains insufficient, the lower priority will
feed the product until the volume is satisfied. In the real-world condition, this method is used
to avoid a penalty or contract underperformance. The function helps assist statements that
meet the conditions listed above. The general Vensim PLE+ equation for this transport train
component is shown in Eq.5.5 below:
11
8*) − G$,-./ :−`01 −[G$,-./ \a ≥ 0,
⎛ ! ! " !," ⎞
⎜ !=1 ⎟
⎜ ⎟
11
Q+,WI = !" $%&' &()& (5.5)
& ⎜ ⎟
`01 −[G$,-./ \a ,
⎜ " !," ⎟
⎜ !=1 ⎟
8*) − G$,-./:
⎝ ! ! ⎠
Where:
Train = Train transport volume of mine brand i (i = 1, …, 11)
i
DR = Desired rate volume of market brand j (j = LC, …., HC)
j
Train = Train transport volume from mine brand i to market brand j
i,j
IF THEN ELSE = Generate the if then else function
MS = Mine stockpile volume from market brand i
i
Σ = sum
j = 1:LC; 2:LMC; 3:MC; 4:MHC; 5:HC
The equation explains that, for example, the mine stockpile for quality x reduced by all
of the more prioritized train transport of quality x is greater or has the same value than the
market brand’s desired rate subtracted by the other more prioritized train transport. The x
mine brand will feed the market based on the desired rate reduced by the other more
prioritized train transport. But, if it is smaller, then the x mine brand will only give their
remaining stock into the market brand. In addition, there will be a percentage introduced to
40
|
Colorado School of Mines
|
the equation in some train transport components. This percentage is intended to keep all of
the mine product reserves stock for other market brands and the forthcoming months.
The transport train is named after their delivery priority, from Train 1 through 3
depending on their capability. Because each mine stockpile has a separate train transport
component, the detailed explanation about each component will be discussed further in the
sub-chapter below.
MP1 Train Transport and Transport for IU Stock
This coal company's lowest quality coal is MP1. This MP1's quality is typically utilized
for internal corporate purposes and as an addition if a certain amount is required for market
brands LC and LMC. Thus, three transport components are created for the MP1 to facilitate
those requirements. The priority order for the MP1 is transporting coal for internal use,
transport to the LC, and lastly to the LMC. Therefore, the equations are following those
prioritizations; the resulting detailed equations shown in Table 5.9 below.
Table 5.9 MP1 Train Transport Equation
MP1 Train Vensim PLE+ Equation
IU Stock MP1 to IU = IF THEN ELSE ((MP1 MS) - (Internal Usage) ≥ 0, Internal
Usage, MP1 MS)
Train 1 MP1 Train 1 = IF THEN ELSE ((MP1 MS-MP1 for IU Stock) - (LC DR-
(to LC) (MP2 Train 1+MP3 Train 1+MP4 Train 2)) ≥ 0, LC DR - (MP2 Train 1 +
MP3 Train 1 + MP4 Train 2), (MP1 MS - MP1 for IU Stock))
Train 2 MP1 Train 2 = IF THEN ELSE ((MP1 MS - (MP1 Train 1 + MP1 for IU
(to LMC) Stock)) - (LMC DR - (MP3 Train 2 + MP4 Train 1 + MP5 Train 1 + MP6
Train 3 + MP2 Train 2 + MP7 Train 3)) ≥ 0, LMC DR - (MP3 Train 2 +
MP4 Train 1 + MP5 Train 1 + MP6 Train 3 + MP2 Train 2 + MP7 Train 3),
MP1 MS-(MP1 Train 1 + MP1 for IU Stock))
Besides from the internal usage that use MP1 exclusively, we can see that in both LC
and LMC quality market brands, MP1 comes in last after all other mine brands' volume is
delivered. The priority is after MP2, MP3, and MP4 in the LC quality. And for LMC, it is in
the order after MP5, MP3, MP2, MP6, MP4, and MP7. But for internal usage, the MP1 is a
single product that supplies this demand.
41
|
Colorado School of Mines
|
MP2 Train Transport
The MP2 is a high-priority product for the LC brand. Besides that, this quality is a
product used to make up for a shortage of LMC. As a result, this product contains only two
train transport components, namely LC and LMC components. In the MP2 train components,
the transport is not utilizing its whole stock into one quality. This is because the MP2 quality
has large production and needs to give balance quantity supplying both market brands. Then
the percentage in the equation is introduced to the components. This is done to ensure that
stock remains available if it is required to meet the LMC market brand's needs. In Table 5.10,
the equation employed in this MP2 train transport is detailed.
Table 5.10 MP2 Train Transport Equation
MP2 Train Vensim PLE+ Equation
Train 1 MP2 Train 1 = IF THEN ELSE ((MP2 MS) - (LC DR) ≥ 0, (LC DR) * 0.9,
(to LC) (MP2 MS) * 0.65)
Train 2 MP2 Train 2 = IF THEN ELSE ((MP2 MS - MP2 Train 1) - (LMC DR-
(to LMC) (MP3 Train 2 + MP5 Train 1 + MP6 Train 3 + MP4 Train 1 + MP7 Train
3)) ≥ 0, (LMC DR-(MP3 Train 2+MP5 Train 1+MP6 Train 3+MP4 Train
1+MP7 Train 3)), (MP2 MS-MP2 Train 1))
The quality of the MP2 utilized in the Train 1 equation is only 90% of the desired rate
if this quality has a larger volume than the desired rate of the market brand LC. Meanwhile, if
the MP2 quality stock is lower, just 65% of the volume will be transported to the LC quality
stockpile port. This MP2 product is only added if there is a lack of volume for delivery to
LMC quality. Therefore, the equation used to satisfy the LMC quality does not employ any
percentage.
MP3 Train Transport
Although it contributes the most volume to the supply of the LMC market brand, MP3
is the second to be prioritized. Besides LMC, MP3 also fits the LC market brand, which
becomes a second priority after the MP2 mine brand. However, the LMC market brand
continues to be given first prioritization for MP3 quality. Table 5.11 contains the complete
equation used for all MP3 train transport.
42
|
Colorado School of Mines
|
Table 5.11 MP3 Train Transport Equation
MP3 Train Vensim PLE+ Equation
Train 1 MP3 Train 1 = IF THEN ELSE ((MP3 MS) - (LC DR - (MP2 Train 1)) ≥ 0,
(to LC) (LC DR - (MP2 Train 1)) * 0.9, (MP3 MS))
Train 2 MP3 Train 2 = IF THEN ELSE ((MP3 MS - MP3 Train 1) - (LMC DR -
(to LMC) MP5 Train 1) ≥ 0, (LMC DR - MP5 Train 1) * 0.8, (MP3 MS - MP3 Train
1) * 0.9)
As previously stated, the quality of this MP3 has priority in delivery to the quality of
LMC coal. However, the MP3 equation was modified to first ensure supply to LC quality by
providing coal intake for the remaining LC desired rate after MP2 and before its transport to
the LMC quality.
In the MP3 Train 1 equation shown, when supplying volume to LC quality, only 90%
of the LC desired rate is given to maintain the stability of the MP3 product volume if the LC
desired rate is lower than the MP3 stock. In the MP3 Train 2 equation for LMC quality, a
percentage of 80% of the desired rate of LMC quality is given if the remaining MP3 products
exceed the desired volume, and 90% of the total MP3 inventory if the volume is less than the
desired rate. It can also be seen that in the delivery of LMC quality, MP3 quality is also
carried out after MP5 quality delivered since MP5 is only transport their whole stock to the
LMC.
MP4 Train Transport
This MP4 train transport's quality differs slightly from the MP3 train transport. The
distinction is in the order of delivery, with MP4 supplying its inventory volume first to the
LMC market brand and subsequently to the LC brand market. After MP5, MP3, and MP6,
MP4 quality is utilized as the third priority in the LC market brand, and it is the fourth
priority in LMC quality. Table 5.12 contains the complete equation for this MP4 transport
train.
43
|
Colorado School of Mines
|
Table 5.12 MP4 Train Transport Equation
MP4 Train Vensim PLE+ Equation
Train 1 MP4 Train 1 = IF THEN ELSE ((MP4 MS) - (LMC DR - (MP3 Train 2 +
(to LMC) MP6 Train 3 + MP5 Train 1)) ≥ 0, (LMC DR - (MP3 Train 2 + MP6 Train 3
+ MP5 Train 1)), (MP4 MS))
Train 2 MP4 Train 2 = IF THEN ELSE ((MP4 MS - MP4 Train 1) - (LC DR -
(to LC) (MP2 Train 1 + MP3 Train 1)) ≥ 0, LC DR - (MP2 Train 1 + MP3 Train 1),
MP4 MS - MP4 Train 1)
Because MP4 quality is a complimentary quality to suit the needs of both market
brands, LMC and LC, the entire equation for MP4 quality does not employ a percentage
share. The MP4 transport to the LMC comes after MP3, MP5, and MP6. Whereas in the LC
market brand, it is in priority number three after MP2 and MP3.
MP5 Train Transport
The MP5 mine product coal only delivers its coal to the LMC market brand. The whole
stock in this MP5 is always spent on the LMC quality. The train transport in the MP5 mine
only has one component to the LMC brand market, which can be seen in Table 5.13 below.
Table 5.13 MP5 Train Transport Equation
MP5 Train Vensim PLE+ Equation
Train 1 MP5 Train 1 = IF THEN ELSE ((MP5 MS) - (LMC DR) ≥ 0, (LMC DR),
(to LMC) MP5 MS)
There is no percentage introduced in the equation, since it will be giving the whole
stock to the market brand. MP5 is also puts in the first order to ensure the whole stock is
delivered to the LMC.
MP6 Train Transport
Train transport for the MP6 product prioritizes the MC market brand. This is because
MP6 has a similar quality to the quality of the MC market brand. However, MP6 also has
transport for other product quality, namely MHC and LMC, to satisfy the needs of both
qualities. So, the priority for MP6 could be sorted as MC, MHC, and LMC. The percentage
for ensuring the distribution proportion is also introduced in the MP6 train transport
components with detailed equations seen in Table 5.14.
44
|
Colorado School of Mines
|
Table 5.14 MP6 Train Transport Equation
MP6 Train Vensim PLE+ Equation
Train 1 MP6 Train 1 = IF THEN ELSE (MP6 MS - (MC DR-MP6 MS) ≥ 0, (MC
(to MC) DR) *0.3, MP6 MS * 0.8)
Train 2 MP6 Train 2 = IF THEN ELSE ((MP6 MS - (MP6 Train 1)) - (MHC Dr -
(to MHC) (MP8 Train 2 + MP10 Train 1)) ≥ 0, (MHC Dr - (MP8 Train 2 + MP10
Train 1)), (MP6 MS - (MP6 Train 1)) * 0.7)
Train 3 MP6 Train 3 = IF THEN ELSE ((MP6 MS - (MP6 Train 1 + MP6 Train 2))
(to LMC) - (LMC DR - (MP3 Train 2 + MP5 Train 1)) ≥ 0, (LMC DR - (MP3 Train 2
+ MP5 Train 1)), (MP6 MS - (MP6 Train 1 + MP6 Train 2)))
The percentage given by MP6 for the MC quality is 30% of the total desired stock if
the available supply exceeds the desired stock of MC quality, and 70% of the total stock of
MP6 if the desired stock exceeds the total stock of MP6. While for MHC and LMC quality,
this percentage of MP6 quality is only used for MHC quality when the desired stock of MHC
is more than the stock owned by MP6 minus delivery to MC quality. This whole percentage
is given to maintain the inventory of MP6 so that it can always supply all quality in the
following months.
MP7 Train Transport
MP7 coal quality is mine quality with train transport components similar to MP6 coal
quality. This quality has three components in order of priority for sending MC, MHC, and
LMC. The complete equation for all MP7 components can be seen in Table 5.15 below.
Table 5.15 MP7 Train Transport Equation
MP7 Train Vensim PLE+ Equation
Train 1 MP7 Train 1 = IF THEN ELSE ((MP7 MS) - (MC DR - (MP6 Train 1)) ≥ 0,
(to MC) (MC DR-(MP6 Train 1)) * 0.6, (MP7 MS) *0.8)
Train 2 MP 7 Train 2 = IF THEN ELSE ((MP7 MS - (MP7 Train 1)) - (MHC Dr -
(to MHC) (MP8 Train 2 + MP10 Train 1 + MP6 Train 2 + MP9 Train 1)) ≥ 0, (MHC
DR - (MP8 Train 2 + MP10 Train 1 + MP6 Train 2 + MP9 Train 1)), (MP7
MS - (MP7 Train 1)))
45
|
Colorado School of Mines
|
Table 5.15 Continued
Train 3 MP7 Train 3 = IF THEN ELSE ((MP7 MS-(MP7 Train 1 + MP7 Train 2)) -
(to LMC) (LMC DR-(MP3 Train 2 + MP4 Train 1 + MP5 Train 1 + MP6 Train 3)) ≥
0, (LMC DR - (MP3 Train 2 + MP4 Train 1 + MP5 Train 1 + MP6 Train 3))
* 0.6, (MP7 MS - (MP7 Train 1)) * 0)
In the train transport component to MHC, MP7 coal uses 60% of the total desired rate
if the coal stock exceeds the desired volume and 80% of the MP7 entire stock if the MP7 coal
stock is below the desired volume. In addition, MP7 coal will allocate all current needs or
stock according to HC and MC coal needs. The exception to LMC quality, if the stock MP7
is less than the desired rate, as this MP7 quality will not supply LMC quality.
MP8 Train Transport
MP8 mine product only has two train transport components, going to MHC and MC.
The priority is given to the MHC market brand because of the similarity in their specification.
After fulfilling the MHC quality, this quality is used to fulfill the MC quality if it has the
remaining volume contained in the stockpile. The complete equation for this MP8 Train
transport can be seen in Table 5.16 below.
Table 5.16 MP8 Train Transport Equation
MP8 Train Vensim PLE+ Equation
Train 1 MP8 Train 1 = IF THEN ELSE ((MP8 MS) - (MHC Dr - (MP10 Train 2)) ≥
(to MHC) 0, (MHC Dr -(MP10 Train 2)) * 0.6, (MP8 MS) * 0.05)
Train 2 MP8 Train 2 = IF THEN ELSE ((MP8 MS - MP8 Train 1) - (MC DR -
(to MC) (MP6 Train 1 + MP7 Train 1 + MP10 Train 3)) ≥ 0 , (MC DR - (MP6 Train
1 + MP7 Train 1 + MP10 Train 3)), (MP8 MS - MP8 Train 1))
At MP8 mine quality, when the stock owned is greater than the remaining desired rate,
60% of the remaining desired rate will be given by this quality. Whereas if the stock of this
mp8 is smaller than the remaining desired rate, then only 5% of the total stock of mp8 is
given to this MHC quality. the rest of this MP8 stock will later be allocated to quality MC.
MP9 Train Transport
This MP9 train transport has the same components and deliveries prioritization as the
MP8 component. The transport train is made to meet MHC and MC quality at this quality.
46
|
Colorado School of Mines
|
The difference between MP9 and MP8 is that MP8 deliveries always have a higher order than
the MP9 quality. detailed equation for MP9 is described in Table 5.17 below.
Table 5.17 MP9 Train Transport Equation
MP9 Train Vensim PLE+ Equation
Train 1 MP9 Train 1 = IF THEN ELSE ((MP9 MS) - (MHC Dr - (MP10 Train 2 +
(to MHC) MP11 Train 2 + MP6 Train 2 + MP8 Train 1)) ≥ 0, (MHC Dr - (MP10 Train
2 + MP11 Train 2 + MP8 Train 1 + MP6 Train 2)) *0.75, (MP9 MS))
Train 2 MP9 Train 2 = IF THEN ELSE ((MP9 MS - MP9 Train 1) - (MC DR -
(to MC) (MP6 Train 1 + MP7 Train 1 + MP10 Train 3 + MP8 Train 2)) ≥ 0, (MC
DR - (MP6 Train 1 + MP7 Train 1 + MP10 Train 3 + MP8 Train 2)) *0.5,
(MP9 MS - MP9 Train 1) * 0)
The MP9 quality acts as complementary to both qualities. For the MHC market brand,
there will be 75% allocation from remaining desired rate if MP9 stock has larger or same
volume as the remaining desired rate, but when it is smaller, then all of the remaining MP9
stock will be delivered to the MHC quality. In fulfilling MC quality, if the desired rate of this
quality has a smaller volume compared to the total remaining stock of MP9 quality, there will
be only 50% of the total desired rate will be delivered from this product, and if it is larger,
then there is no coal deliveries will be made for MC quality.
MP10 Train Transport
The MP10 has different components than MP8 and MP9 mine quality. The quality of
this MP10 has three transport components that lead to the market brand HC, MHC, and MC.
In the HC component, the MP10 product is only an addition if the MP11 product is
insufficient, while the MP10 product is the priority product for the MHC component.
However, because HC products are premium, delivery priority is still given to HC products.
On the other hand, MP10 is coal with the second to last priority in MC quality. Thus, all
remaining volume from the MP10 stockpile will be delivered to MC quality to fulfill its
desired rate as shown in Table 5.18 below.
47
|
Colorado School of Mines
|
Table 5.18 MP10 Train Transport Equation
MP10 Train Vensim PLE+ Equation
Train 1 MP10 Train 1 = IF THEN ELSE (MP10 MS - (HC DR - MP11 Train 1) ≥ 0,
(to HC) HC DR - MP11 Train 1, 0)
Train 2 MP10 Train 2 = IF THEN ELSE ((MP10 MS - MP10 Train 1) - (MHC Dr)
to MHC ≥ 0, (MHC Dr) * 0.5, (MP10 MS) * 0.5)
Train 3 MP10 Train 3 = IF THEN ELSE ((MP10 MS - MP10 Train 1) - (MC DR -
(to MC) (MP6 Train 1 + MP7 Train 1 + MP8 Train 2)) ≥ 0, (MC DR - (MP6 Train 1
+ MP7 Train 1 + MP8 Train 2)), (MP10 MS - MP10 Train 1))
On the train transport to MHC quality, the percentage used is 50% of the total desired
rate if the stock reserves at MP10 quality exceed the desired rate, and the same percentage of
50% of the total stock MP10 if the desired rate has a more significant volume compared to
the rest of the stock MP10.
MP11 Train Transport
MP11 train transport has similar component with the MP10 but, the MP11 is the main
priority for the HC brand market. In addition, this MP11 can also meet two other market
brand qualities, MHC and MC. So, the number of components in the MP11 train transport is
three components with a equation that can be seen in Table 5.19.
Table 5.19 MP11 Train Transport Equation
MP11 Train Vensim PLE+ Equation
Train 1 MP11 Train 1 = IF THEN ELSE (MP11 MS-HC DR ≥ 0, HC DR , MP11
(to HC) MS )
Train 2 MP11 Train 2 = IF THEN ELSE ((MP11 MS - (MP11 Train 1)) - (MHC Dr
(to MHC) - (MP10 Train 2+MP8 Train 1)) ≥ 0, (MHC Dr - (MP10 Train 2 + MP8
Train 1)) * 0.75, (MP11 MS - (MP11 Train 1)) * 0)
TT to MC MP11 Train 3 = IF THEN ELSE ((MP11 MS - (MP11 Train 1 + MP11
Train 2)) - (MC DR - (MP6 Train 1 + MP7 Train 1 + MP10 Train 2 + MP8
Train 2 + MP9 Train 2)) ≥ 0, (MC DR - (MP6 Train 1 + MP7 Train 1 +
MP10 Train 2 + MP8 Train 2 + MP9 Train 2)), (MP11 MS - (MP11 Train 1
+ MP11 Train 2)))
48
|
Colorado School of Mines
|
U4 = ef QgPh PSTP (T,%(# −1T > 0,T,%(# −1T ,0) (5.7)
- - - - -
Where:
DR = Desired rate volume of market brand j (j = LC, …., HC)
j
PS = Port stockpile volume of market brand j
j
Sales =Sales volume of market brand j
j
IF THEN ELSE = Generate the if then else function
j = 1:LC; 2:LMC; 3:MC; 4:MHC; 5:HC
As previously stated, the tonnage of sales and stockpile ports varies by market brand.
As a result, every market brand has this component variable. Table 5.21 contains the equation
for each component in detail.
Table 5.21 Desired Rate Equation
Market Brand Name Vensim PLE+ Equation
LC IF THEN ELSE (LC Sales-LC Port Stockpile>0, LC Sales-LC Port
Stockpile, 0)
LMC IF THEN ELSE (LMC Sales-LMC Port Stockpile>0, LMC Sales-
LMC Port Stockpile, 0)
MC IF THEN ELSE (MC Sales-MC Port Stockpile>0, MC Sales, 0)
MHC IF THEN ELSE (MHC Sales-MHC Port Stockpile>0, MHC Sales-
MHC Port Stockpile, 0)
HC IF THEN ELSE (HC Sales-HC Port Stockpile>0, HC Sales-HC
Port Stockpile, 0)
The detailed equation in Table 5.21 shows us that every market brand has the same
equation which will be generating value when there is a bigger value from sales compared to
the stockpile availability. This desired rate component then used in the train transport
components as a based value for their coal supply to the market brand.
Desired Stock
This desired stock component is given in addition to the desired rate for components
with low market brand quality (LC, LMC, and MC) to ensure there is stock at the port for
these three components. This is done to reserve stock at the port, should it be needed. This
desired stock is only given to the three components with the lowest quality because due to the
condition of the company that has the most significant sales of these three qualities. While for
higher quality, due to limited reserves, and predictable demand, delivery will only be made
following the current desired rate.
51
|
Colorado School of Mines
|
CHAPTER 6
MODELING AND DEFINING SCENARIO
The next process of this research is a modeling process and defining the scenarios. The
modeling process is carried out in two different stages. The first stage is creating a causal
loop diagram (CLD) in which components are designed and connected according to the
possibilities. After making this causal loop diagram, the modeling process continue with
building a stock and flow diagrams (SFD). in this step, the model is made more detailed and
includes all the equations prepared previously.
Meanwhile, the defining scenario stage is carried out in parallel with model creation.
This is done because of the stages in making this scenario, additional modifications to the
existing equation are needed to be introduced. Introduction to required data and new
parameters are also being done to help describe the scenarios. If all scenarios have been
formed and combined, the model will be re-analyzed to ensure the model run perfectly
without any error.
6.1 Modeling
The modeling process is created following the transport and blending process to fulfill
monthly sales tonnage requirements. The basic concept of the model is when the mine’s coal
quantity of one product has a smaller quantity than the market product, the system will try to
get the coal from other mine quality sequentially, following their nearest quality to the
farthest quality.
If some sales are not fulfilled, the model will show an error in the specific quality and
month to warn of the result. All modeling stages are carried out based on the theory above
and are expected to adapt to actual conditions. There are two modeling processes: a causal
loop diagram to ensure all connections to the model are drawn correctly; and a stock and flow
diagram that can later verify all components and the relationships between components that
have been made.
After creating both CLD and SFD model, it shows that the system dynamics can model
the coal industry’s multi-product supply chain process. After the analysis is done, the system
is also shown the behavior between each component as a graph, and their remaining value as
information regarding the company’s remaining coal inventory. These models help company
to test results of various scenarios, to estimate faster and accurate production and sales, to
make better plans as the system dynamics modeling provide detailed information about the
time dependent behavior of the supply chain.
52
|
Colorado School of Mines
|
11 components. As a result, the total value of the company’s production is divided, allowing
modeling and simulation to see the output at various quality levels. In this example, the
production component is described as a flow component that goes to a mine stockpile per
quality. The mine product component, transformed into a mine stockpile, is the next. As
previously stated, the mine stockpile component in this SFD transforms into a stock
component because a stockpile is a component with an accumulation function. As a result,
this stock component is ideally used for mine stockpile components.
A connection is also introduced to the mine stockpile component that points to the
stockpile market brand component, and this transport component is a link component. The
market brand stockpile component (port stockpile) is also changed into a stock component
since it is a component that requires an accumulation component (as it is also a stockpile
component). Back to the train transport component, this one is used as a connector
component in this SFD model because it connects the two stocks. All these transport
components have the same relationship as the previous components in the CLD, except they
are now connector components.
After that, a sales component transforms into an output with a cloud connector, which
is the model’s limit. Following the company’s market brand, the sales component is created
into five components consist of all the market brand. Domestic and export sales for each
component are presented following market brand sales, so these sales are divided into the
market brand at the time of data analysis.
In the sub-chapter below, a discussion about the complete stock and flow diagram from
each market brand will be detailed. This explanation is done because each market brand is the
end/output of each model, so we can see how the system works from input to output.
6.1.2.1 IU Stock and Flow Diagram
Mine brands that can meet internal usage needs only comprise one mine brand, namely
MP1. This MP1 brand mine also considers IU a top priority for its coal shipments, followed
by priority shipments to the LC and LMC market brands.
55
|
Colorado School of Mines
|
Figure 6.3 LC Stock and Flow Diagram
This effect happens because MP4 Train 2 is the second transport priority on the MP4
mine brand after MP4 Train 1. This shadow variable also can be looked at the MP1 mine
brand. There is a MP1 for IU which affecting the MP1 Train 1 shown as a shadow variable.
on the overall model, this shadow variable does not appear because the arrow will come
directly from MP4 Train 1, as shown in appendix B.2.
6.1.2.3 LMC Stock and Flow Diagram
The fulfillment of LMC market brand can be carried out by seven different qualities:
the quality from MP1 to MP7. As stated in chapter 5.1.3, the priority quality of the LMC
market brand is MP3 quality. However, because MP5 quality uses all of its volumes to fulfill
the LMC market brand, MP5 quality is made to the first quality that will meet the mine
brand. Figure 6.4 shows that MP5 quality gives an arrow to all other mine brands, including
MP3 quality. The arrow has been adjusted according to the priority scale of fulfilling the
LMC market brand.
Figure 6.4 also shows the priority of transport components in this LMC market brand.
Like the LC market brand, all transport components of higher importance give effect through
arrows to transport components with lower priority. Because many qualities are not the main
priority in fulfilling this LMC quality, many priority components provide arrows for the
transport train component to this LMC shown as the shadow variables.
57
|
Colorado School of Mines
|
Figure 6.4 LMC Stock and Flow Diagram
All priority components are usually numbered 1 on the train transport component. so it
can be seen that if there is a component with a large number in their train component, then
that component will get the arrow (effect) from the train transport component with smaller
number.
6.1.2.4 MC Stock and Flow Diagram
Mine brands that can provide the volume of coal to fulfill MC quality ranging from
MP6 to MP11. The priority mine brand for MC quality is MP6. Therefore, Figure 6.5 shows
MP6 effecting through arrows on every other quality, followed by other brand based on their
prioritization.
As we can see, LC is also not a priority for most of its possible mine brand distributor.
Only MP6 and MP7 consider MC as their priority brand. therefore, the train transport
component other than those two components is getting an arrow from other train transport
component. The arrow for the train transport component is also works similar to the arrows in
the LC and LMC quality, which gives an information about the volume that being distributed
every month.
58
|
Colorado School of Mines
|
It can be seen in Figure 6.6 that MP10 has an arrow to other mine brands followed by
mine brands. The process continues same with other market brand. arrows are introduced
from the higher priority to the lower priority mine brands. Shadow variables also introduced
to the mine brand with higher number of train transport components from other market brand
to ensure the exact volume is obtained.
6.1.2.6 HC Stock and Flow Diagram
MP11 and MP10 is the only product that able to feed the HC market brand. These two
mine brands consider the HC market brand as their priority distribution. Therefore, on the
train transport component, there is no effect from other market brand train transport
component. However, it can be seen in appendix B.2, these two products train transport
giving the arrows to the other market brand train transport (MC and MHC) because they have
the possibility of fulfilling volumes for both market brands.
Figure 6.7 HC Stock and Flow Diagram
It can be seen in Figure 6.7 that MP11 gives an arrow to the MP10 train transport
because MP11 is the main quality mine brand for this HC market brand. apart from that, this
component has a similar relationship with other components.
6.2 Defining Scenario
In this study, three different scenarios are introduced, each of them are then tested on
the main model. These three scenarios were chosen based on the possibility of occurrence for
this company. The scenarios introduced into the model are diversification scenario (Scenario
1), and two of the long-term plan supply-demand scenarios which are 2022 supply-demand
scenario (Scenario 2), and 2023 supply-demand scenario (Scenario 3) each of the scenario
will be explained in the next sub-chapter.
60
|
Colorado School of Mines
|
6.2.1 Scenario 1: Diversification Scenario
The diversification scenario was introduced into the model following the company's
long-term plan to diversify and expand their business portfolio, namely the sale of coal
gasification. In this scenario, a new business process will be created according to the
company's calculations and estimates in determining the amount of coal needed. Additional
data carried out in this process will be on the total target of the final product that the company
wants to produce for this business diversification.
In this diversification scenario, the condition is that the company will increase the
supply of low-quality coal to factories that have a total demand for coal gasification products
of 1.2 million t/y. It is known that the total product can be produced from the combustion
process of low-quality coal. From the company’s calculation results, every ton of low-quality
coal combustion is estimated to produce 0.2 tonnes of coal gasification products which
represents a 20% recovery rate.
With this fact, two new stocks components will be introduced to the main model
containing the total target of the final product diversification, in this case, coal to gas
inventory (CTG inventory) and a stockpile for storing coal before the combustion process
(coal for CTG). The stocks are connected by the flow components, namely the burning
process, which feeds the CTG inventory from the coal for CTG based on the CTG desired
rate variable. The CTG desired rate is getting its value from the distribution flow, which acts
as an output from CTG inventory compared to the total available CTG inventory.
If the CTG inventory is smaller than the distribution demand, the CTG desired rate will
request the burning process to take coal from the coal for CTG flow and transfer it into the
CTG inventory to satisfy the demand. Distribution flow will have a value of total coal
gasification needed each month. The value will be generated by the company and put into the
database. In this research, this study is using the 100,000 tonnes per month generated from
total yearly demand of 1.2 Mt. In this scenario, as we can see from Figure 6.8, to connect the
initial model with the scenario, additional flow is introduced connecting the mine stockpile of
low-quality coal (MP1, MP2 and MP3) to the coal to CTG stock.
61
|
Colorado School of Mines
|
Figure 6.8 The Diversification Scenario Illustration
A variable containing the value of the coal recovery rate into a gas recovery rate will
also be introduced as a parameter for the equation in the CTG inventory and the CTG desired
rate. The recovery rate is used to adjust the tonnage of coal from coal to CTG to the CTG
inventory due to the burning process/combustion. Conversely, it will act as the multiplier for
the demand for coal gasification in the tonnage of raw materials.
The flow tool here acts as a transport process to the stockpile before the combustion
process occurs. Flow tools will also be introduced as the combustion process occurs between
the final stockpile and coal gasification. The flow tools are affected by previous transport
flow in each mine product. The affecting flows for the MP1 to CTG stockpile flow are the
MP1 for IU stock, Train 1 and Train 2. For the MP2 to CTG stockpile flow, the affecting
flows are Train 1 and Train 2. Then, for the MP3 to CTG, the affecting parameters are Train
1 and Train 2. The MP1 to CTG will also affect the MP2 and MP3 to CTG flow since the
MP1 will be the priority product to feed the coal gasification product.
If the MP1 product cannot satisfy the desired rate, then the MP2 product will start to
provide the coal for CTG flow. The process is continued to MP3 if the MP2 also cannot
fulfill the demand of CTG. With the added flow for three mine brands, the equation of them
will also added by the CTG transport. The added equation can be seen in Table 6.1 below.
Table 6.1 Mine stockpile component modification for diversification scenario
Component Name Vensim PLE+ Equation
MP1 MS MP1 MS = (MP1 Prod)-MP1 Train 1-MP1 Train 2-MP1 for IU
Stock-MP1 to CTG
MP2 MS MP2 MS = (MP2 Prod)-MP2 Train 2-MP2 Train 1-MP2 to CTG
MP3 MS MP3 MS = (MP3 Prod)-MP3 Train 2-MP3 Train 1-MP3 to CTG
62
|
Colorado School of Mines
|
The CTG transport will act as the last priority in every mine product. As we can see,
the equation added the CTG in the previous parameters to reduce the value of products being
stored in the stockpile. Other components that being added in this diversification scenario
also given the equation which similar to the equation used in the main model component. The
added components equation could be seen in Table 6.2 below.
Table 6.2 Added component equation for Scenario 1
Component Name Equation
Coal for CTG Coal for CTG = (MP1 to CTG + MP2 to CTG + MP3 to CTG) -
Burning Process
CTG Inventory CTG Inventory = (Burning Process*Recovery Rate)-Distribution
CTG Desired Rate CTG DR = IF THEN ELSE (Distribution - CTG Inventory > 0,
(((Distribution - CTG Inventory)) / (Recovery Rate)), 0)
In the Coal for CTG Stock, the equation used will be like the port stockpile component,
consisting of total coal delivered from the mine stockpile, reduced by the need burning
process, which gets its value from the total desired rate. Stock component of CTG inventory
also has similarities with the component of coal for CTG. The difference lies in the input
from the burning process, which is further reduced by distribution to gas factories. In this
component, the value generated is already coal in the form of gas because it has gone through
the combustion process. As previously explained, this inventory will describe the results of
the burning process multiplied by the recovery rate, which will result in a reduction in
volume. Therefore, the equation used for the desired rate is also considered so that the
conversion between gas and coal is equally aligned and does not cause conversion errors.
From the CTG Desired Rate equation, it can be concluded that the actual value of coal needs
is not in the form of gas, so when requesting a value for all mine products, the requested
value is in the form of coal, not in the form of gas.
The purpose of this scenario is to determine the amount of volume needed to be able to
run a diversified business without disrupting the existing system. In this scenario, several
simulations are carried out to produce a minimum volume to meet the volume requirements
of this CTG manufacture process.
63
|
Colorado School of Mines
|
6.2.2 Supply and Demand Scenario
The supply and demand scenarios will be the second and third scenarios given in the
model. In the supply and demand scenario, two different scenarios will be analyzed, namely
the scenario for 2022 and 2023. The difference between these two scenarios lies in the
percentage increase in the company's production and sales so that it will later produce
different outputs for each component.
The purpose of this scenario is to see how the behavior of the existing system changes
when using different volumes. In this scenario, it can also be seen how the company's
inventory is every year. further analysis can be carried out that can assist the company in
overcoming the existing problems from the results of this scenario simulation.
Before the analysis of these two scenarios is carried out, it is necessary to plan the
components that will be made to accommodate the changes in the percentage for production
and sales. These component plans and equation are discussed in the next sub-chapter.
6.2.2.1 Component of the Supply and Demand Scenario
Two components are presented in the supply and demand scenario and this component
is the production rate component and the sales rate component. These two components are
included in each appropriate mine brand and market brand so that they can be modified
according to the company's long-term plan.
Production Rate Component
The component incorporated into the earlier model is the production rate component
created by adding one component for each quality; this component is called the production
rate. The equation consists of the percentage of production for each grade from the company's
long-term plan. This parameter will use an auxiliary variable, which will allow game
simulations in the future.
Each parameter of the production rate will have a slider that can alter the production
percentage of each quality, with the effects viewable right away. It is set to a value of 0-3,
with 0 denoting no production, 1 denoting 100% performance, 2 denoting 200% of
production and 3 denoting 300% of production. The range of the slider is set to 0.1, which
means that it will move for each 10% rise or drop in production, depending on the simulation.
The Figure 6.9 shows how the production rates component is introduced to the model.
64
|
Colorado School of Mines
|
Figure 6.9 The Production Rate Illustration
This production rate component is added for each mine brand in the actual model. After
being linked to the production rate component, equation modifications are made to the mine
stockpile component. To consider the production rate component in the simulation to be
Simulation. After adding the production rate, the basic equation utilized in the mine stockpile
component is shown in the Eq. 6.1:
ZT = (1+$2 ; 14 )−8GQ+,WI : (6.1)
& & & &
Where:
MS = Mine stockpile volume of mine product i (i = 1,…,11)
i
Prod = Production volume of mine product i
i
PR = Production rate of mine product i
i
Σ = sum
Train = Train transport volume from mine product i
i
The equation added the production rate as a multiplication for the production to
produce the production volume matches company’s long-term plan. Furthermore, this
fundamental equation is applied to all components of the mine stockpile, as shown in Table
6.3:
Table 6.3 Production Rate Equation Modification
Mine Product Name Equation
MP1 (MP1 Prod*MP1 PR)-MP1 Train 1-MP1 Train 2-MP1 for IU
Stock-MP1 to CTG
MP2 (MP2 Prod*MP2 PR)-MP2 Train 2-MP2 Train 1-MP2 to CTG
MP3 (MP3 Prod*MP3 PR)-MP3 Train 2-MP3 Train 1-MP3 to CTG
MP4 (MP4 Prod*MP4 PR)-MP4 Train 1-MP4 Train 2
MP5 (MP5 Prod*MP5 PR)-MP5 Train 1
65
|
Colorado School of Mines
|
Table 6.3 Continued
MP6 (MP6 Prod*MP6 PR)-MP6 Train 1-MP6 Train 2-MP6 Train 3
MP7 (MP7 Prod*MP7 PR)-MP7 Train 2-MP7 Train 1-MP7 Train 3
MP8 (MP8 Prod*MP8 PR)-MP8 Train 1-MP8 Train 2
MP9 (MP9 Prod*MP9 PR)-MP9 Train 2-MP9 Train 1
MP10 (MP10 Prod*MP10 PR)-MP10 Train 1-MP10 Train 2
MP11 (MP 11 Prod*MP11 PR)-MP11 Train 1-MP11 Train 2-MP11 Train
3
We can see that the modification is made on the production component from all the
equations above. The production component will be multiplied by the component production
rate so that all production data will produce output according to changes in production in the
desired year.
Sales Rate Component
The sales rate component will also be introduced to the model in the supply-demand
scenario, which is comparable to the production rate component. The components utilized are
variable components with the same aim as the production components: to allow modeling to
be executed according to the desired percentage. The supply rate component employs the
same template value as the production component, namely 0-3, with 0 representing no sales
and 3 representing 300% sales. The example utilized in Figure 6.10 is similar to the
production component, however the side that is visible is the sales side
Figure 6.10 The Sales Rate Component Illustration
. In the first case, the percentage of maximum sales that may be created from the same
production volume as the existing data will be determined. This sales rate component will
66
|
Colorado School of Mines
|
change the sales equation in the initial model, so that the sales equation will be shown in
eq.6.2 below:
T,%(# = T4 ; i PT +UT j (6.2)
- - - -
Where:
Sales = Sales volume of market brand j (j = LC, LMC, MC, MHC, HC)
j
SR = Sales rate of market brand j
j
ES = Export sales volume of market brand j
j
DS = Domestic sales volume of market brand j
j
As we can see from Eq.5.2, the sales rate will become a multiplication to the total of
domestic and export sales. the multiplication will result the total sales with changes depends
on the percentage given. Following the acquisition of the basic equation, all sales components
are updated and adjusted to this basic equation, resulting in the comprehensive equation
shown in Table 6.4
Table 6.4 Demand Scenario Equation
Mine Product Name Equation
LC Sales LC SR*(LC Dom+LC Ex)
LMC Sales LMC SR*(LMC Dom+LMC Ex)
MC Sales MC SR*(MC Dom+MC Ex)
MHC Sales MHC SR*(MHC Dom+MHC Ex)
HC Sales HC SR*(HC Dom+HC Ex)
This scenario's long-term data comes from the same source as the previous supply
scenario. The money data used, on the other hand, is just data on the likelihood of a growth in
the value of sales, as shown in Table 6.4.
6.2.2.2 Scenario 2: 2022 Supply and Demand Scenario
In the second scenario, the company's long-term plan is used in 2022. In this company's
strategy, the company has determined the increase in sales and production. The data provided
by the company is then input into the equation that was made previously. There are data
limitations in the company data taken where all sales are not divided based on quality but
only into exports and imports. Likewise, the data provided for production is only in the
percentage increase in each existing production field. In this scenario, the company's
predetermined percentage of production and sales for each quality is used. Table 6.5
describes the percentage of production and sales that will be added in 2022.
67
|
Colorado School of Mines
|
Table 6.5 Percentage change from 2021 Production for 2022 Supply-Demand Scenario
Mine Product % Change from 2021
MP1 60%
MP2 91%
MP3 60%
MP4 139%
MP5 136%
MP6 163%
MP7 109%
MP8 109%
MP9 91%
MP10 136%
MP11 109%
Table above explains the production and sales have increased and decreased in the
company's predictions for 2022. in the quality of MP1, MP2, MP3, and MP9, there was a
decrease in coal production, while other qualities experienced an increase production
compared to 2021. Table 6.6 shows about the percentage change in the sales sector from
2021 that will be used for this scenario.
Table 6.6 Percentage change from 2021 sales for 2022 Supply-Demand Scenario
Sales Type % Change from 2021
Domestic Sales 126%
Export Sales 83%
Domestic sales will increase by 26% from the sales level in 2021. While for export
sales, the percentage will decrease with total sales of only 83% compared to 2021.
6.2.2.3 Scenario 3: 2023 Supply and Demand Scenario
The supply and demand scenario for 2023 uses the same analysis methodology as the
2022 scenario. The difference between this scenario and the last lies only in the percentages
used. Long-term company data for 2023 is used and processed the same way as the 2022
scenario, resulting in changes in production and sale component. The percentage used in
2023 is a percentage compared to 2021, which is the same comparison as used in 2022
scenario. This is done because the primary data used is data for 2021, so the percentage
68
|
Colorado School of Mines
|
change is compared to the preliminary data. In addition to a comparison with the value for
2021, the analysis carried out also compares 2022 output, allowing the differences for each
year to later be seen. In table 6.7, we can see the percentage used in this scenario in 2023. the
majority of the production percentage in 2023 is smaller than the 2022 scenario.
Table 6.7 Percentage change from 2021 Production for 2023 Supply-Demand Scenario
Mine Product % Change from 2021
MP1 46%
MP2 92%
MP3 46%
MP4 113%
MP5 113%
MP6 163%
MP7 91%
MP8 73%
MP9 92%
MP10 113%
MP11 92%
Only the MP6 mine brand has the same percentage of production in this scenario and
the percentage increases in MP2 and MP9. The company does this to maintain the amount of
stock available at the end of 2023 in accordance with the prediction of remaining stock in
2022. Table 6.8 shows about the percentage change in the sales sector from 2021 that will be
used for this scenario.
Table 6.8 Percentage change from 2021 Sales for 2023 Supply-Demand Scenario
Sales Type % Change from 2021
Domestic Sales 137%
Export Sales 62%
Domestic sales will increase by 37% from the sales level in 2021. while for export
sales, the percentage will decrease with total sales of only 62% compared to 2021.
69
|
Colorado School of Mines
|
In calculating the accuracy with the precision accuracy method, the total model
accuracy reached 73%. This can be interpreted that from all analyzes carried out by the
model, the accuracy obtained is only 73%. Furthermore, in the RSME method, the overall
RSME value obtained is 65,365. This overall value is obtained because the range of
prediction errors ranges from tens to hundreds of thousands of tonnes. In the last accuracy
calculation method using MAPE, the overall error percentage value in the model is 27%. This
illustrates that there are still around 27% errors that occur in the model's prediction.
In the calculation, all accuracy is calculated based on data per transport component and
calculated monthly. Furthermore, from 12 months of data, an average estimate is carried out
that describes the overall accuracy of the component. Then, the overall accuracy of all
models is calculated based on the overalls of all existing components. Detailed data for the
calculation of MAPE, RSME and Absolute Accuracy for Transport component could be seen
in appendix E.1, E.2, E.3, Respectively.
After all the accuracy calculations have been obtained, the ranking is carried out for all
these transport components. This is done to see how the level of confidence in the
components is determined and the confidence rating is determined. In Table 7.1, we can see
the ranking order of all transport components in the system.
Table 7.1 Accuracy Rank for Transport Component
Mine Product MAPE RSME ABSOLUTE
MP1 for IU Stock 1 1 1
MP1 Train 1 1 1 1
MP1 Train 2 1 1 1
MP5 Train 1 2 3 2
MP11 Train 1 3 2 3
MP7 Train 3 4 4 4
MP4 Train 2 5 11 5
MP10 Train 1 5 2 5
MP3 Train 2 6 21 6
MP2 Train 1 7 22 7
MP4 Train 1 8 17 8
MP10 Train 2 9 5 9
MP7 Train 1 10 18 10
71
|
Colorado School of Mines
|
49.95% for MAPE calculation. The result of those three methods is then calculated and given
the rank. Combined with the transport validation rank, the rank from this stock validation is
then used as a reference for further analysis. The results of the ranking of all mine stocks are
listed in Table 7.3 below.
Table 7.2 Stock Validation Rank
Mine Product MAPE RSME ABSOLUTE
MP1 1 1 1
MP5 1 1 1
MP10 2 2 2
MP9 3 8 3
MP3 4 9 4
MP11 6 6 6
MP2 5 10 5
MP4 9 3 9
MP7 7 7 7
MP8 8 5 8
MP6 10 4 10
The MP1 and MP5 mine products are still getting the first rank in this stock validation
ranking. In addition, the MP10 mine brand is getting the second rank because the stock
distribution similarity resembles its actual conditions. On the other hand, the MP8 and MP6
have the worst accuracy compared with their actual stock condition. As previously stated,
products that have priority in coal shipments in the middle order tend to have a lower rank
than priority products from each market brand.
73
|
Colorado School of Mines
|
Figure 8.1 0 Initial Inventory Stock Simulation Result
This happens because all mine brands that can supply CTG have no sufficient stock to
provide the volume of coal to this CTG component. Components MP1, MP2, and MP3
supply the coal at TIME 1, but when combined, the total volume supplied is not sufficient to
meet the desired stable rate of 500,000 tonnes per month.
The figure also shows that at TIME 5, all new coal supplies can meet the CTG's desired
rate. However, the volume shortage has reached 1,500,000 tonnes accumulation from the
third to the fifth month. In the eighth month, although the MP1 mine brand cannot meet the
desired rate of this CTG, the MP2 and MP3 mine brands already have sufficient inventory to
provide supply to CTG.
These results conclude that if the company does not have coal reserves allocated
specifically for CTG, CTG production will only last for one month. The LC and LMC will
also have stock deficiencies because the MP1, MP2, and MP3, are among the main qualities
in fulfilling those two market brands. From the coal volume perspective, there is a shortage of
1.5 Mt of coal, which means CTG production is not possible for three months in months 3, 4,
and 5.
8.1.2 Simulation with 1,5 Mt Inventories
Based on the deficit volume in the simulation without any CTG inventory, the second
simulation with the 1,5 Mt initial inventory is conducted. The simulation runs without errors
75
|
Colorado School of Mines
|
in both the main model and the scenario component. The behavior of the simulation result
can be seen from Figure 8.2.
Diversification : 1.5 Mt Inventories
1,600,000.000
1,400,000.000
1,200,000.000
1,000,000.000
800,000.000
600,000.000
400,000.000
200,000.000
-
1 2 3 4 5 6 7 8 9 10 11 12
Month
MP1 to CTG : CTG 1,5m MP2 to CTG : CTG 1,5m MP3 to CTG : CTG 1,5m
CTG Desired Rate : CTG 1,5m Coal for CTG : CTG 1,5m CTG Inventory : CTG 1,5m
Figure 8.2 1.5 Mt Initial Inventory Stock Simulation Result
From Figure 8.2, it is shown that most of the components do not indicate any
movement in the entire year. Within the scenario, we can only see a degradation movement
from the CTG inventory component to 400,000 tonnes of coal at the end of the year.
Therefore, there is 400,000 tonnes of extra excess stock if the company is going to dedicate
1,5 Mt inventory for this scenario at the beginning of the year.
8.1.3 Simulation with 400,000 Inventories
The third simulation uses the volume based on the excess inventory in the second
simulation, which is 400,000 tonnes for the initial inventory. This simulation shows that the
scenario component has some movement in the entire year, as seen in Figure 8.3.
76
)snoT(
emuloV
|
Colorado School of Mines
|
Figure 8.4 500,000 Initial Inventory Stock Simulation Result
Figure 8.4 shows similar behavior with the 400,000 tonnes simulation. The difference
lay on the fact that the desired rate is achieved at increasing starting point, where it starts in
the month 6 rather than month 5 like the previous scenario. This difference creates an effect
for other components to have 1 month difference as compared to the fourth scenario. The
result from this extra one month is due to the main system no longer disturbed by the scenario
which means that there is enough coal to run the scenario with 500,000 tonnes of coal.
8.1.5 Simulation with Exact Inventory
Although the 500,000 tonnes inventory simulation already satisfies the demand for
CTG, the search for the exact CTG initial inventory value needs to be done to know the
minimum inventory needed for running the scenario without disturbing the primary model.
The simulation was conducted by trial and error between the fourth and fifth simulation.
Then, it was discovered that a volume of 448,158 tonnes of coal is the minimum inventory
value for the component to run the scenario with enough inventory without disturb the main
model. After using simulation, it can be seen in Figure 8.5 that the behavior obtained from
the initial volume of 448,158 tonnes of coal.
78
|
Colorado School of Mines
|
Figure 8.5 Exact (448,158 tonnes) Initial Inventory Stock Simulation Result
In this simulation, the desired rate starts at TIME 5 with value of 259,210 tonnes. This
is caused by the remaining inventory in the CTG inventory component, which the MP1 to
CTG component provides deficiencies starting at TIME 5 for this component. This difference
then affects the total availability of the MP1 component, which runs out at TIME 11, making
the intake purely came from the MP2 and MP3 products without using MP1 products.
The MP1 mine brand in this simulation is giving the total of 2.7 Mt of coal which is
similar with previous simulation. While for the MP2 mine brand, the total intake for CTG is
565,093 tonnes and for MP3 mine brand is 434,906 tonnes of coal. It is concluded that this
simulation has an adequate supply of coal for the CTG. This simulation also does not
interfere with the initial model coal distribution.
8.2 Scenario 2 (Supply – Demand 2022) Result
The second scenario carried out in this research is the supply and demand scenario of
the company's 2022 plan. The percentage value in the Table 5.5 is used in the demand rate
and supply rate component. In this scenario, the remaining inventory in 2021 is used as initial
inventory for each mine brand which can be seen in table 8.1. This is done with the
assumption that this 2022 supply-demand scenario runs after the model in 2021 has been
successfully simulated.
79
|
Colorado School of Mines
|
CHAPTER 9
CONCLUSION AND RECOMMENDATION
In this chapter, the research questions are answered based on the analysis that have
been done. In addition, solutions are also given to problems found from the analysis so that
the company will know the existing shortcomings. Overall, this study proves it is possible to
make a decision support tool in the coal industry supply chain using the system dynamics
method.
9.1 Conclusion
1. The system dynamics can model the coal industry’s multi-product supply chain process
with acceptable accuracy. After the analysis is done, the system behavior can also be
analyzed for each component as a graph. Moreover, information regarding the company’s
remaining stock of coal and predicting future possible condition can be obtained. This
helps the company to have a decision on what product to be their focus to execute their
future sales and business developments. Additionally, with the system dynamics model,
the company can increase their ability to create faster and accurate production, sales, and
future scenario planning.
2. The accuracy calculated in this study is the comparison between the model's output
results with the values that exist in actual conditions. The calculated accuracy values are
the accuracy of transport accuracy and the accuracy of the remaining stock.
The accuracy of the transport component is calculated using three different calculation
methods, namely Absolute accuracy, RMSE, and MAPE. The accuracy value generated
from the absolute accuracy calculation is 73%, RMSE is 65,365, and for MAPE, it is
27%.
For the stock component, same accuracy calculation was also carried out and resulted in
an overall accuracy value for this component of 50.04% for absolute accuracy, 115,220
for RMSE calculation, and 49.95% for MAPE calculation.
3. Three scenarios are introduced to the model; Scenario 1: diversification scenario, where
the company tries to create coal to gas product; Scenario 2: 2022 supply-demand
scenario; and scenario 3: 2023 supply-demand scenario. All scenarios were using the
company's long-term data and created based on the possibility of occurrence for this
company.
In the first scenario, three different simulations with different initial inventory are
performed. The first simulation assumes that the company does not have any stock for
83
|
Colorado School of Mines
|
this scenario. From this calculation, it is found that the company will stop production at
TIME 3, 4, and 5 because it does not have enough stock to supply the needs of this
scenario. The second simulation assumes the company has 500,000 tonnes of initial
inventory, and this calculation shows that the company will have excess stock because the
initial inventory is too large. In the third simulation, the exact initial inventory volume is
provided so that the company can meet the needs of the CTG demand. It was found that
the inventory needed 448,158 tonnes of initial inventory as the minimum inventory
requirement to carry out this scenario in 2021.
The second scenario is the supply-demand scenario in 2022. The changes in supply and
demand are introduce according to the company's plan for 2022. It can be concluded that
in this scenario, there is no shortage of coal supply from the changed production level.
The remaining inventory for 2022 Supply-Demand scenario is 5.2 Mt of coal.
The third scenario is the scenario of changes in supply and demand in 2023. There is also
no shortage of coal supply from the changed production level in this scenario. In the
scenario 3, the remaining inventory is 5.5 Mt.
From the scenario 2 and 3 it is concluded that the company is trying to maintain its stock
at the end of the year, seen from the similarity of the total stock obtained from the model,
scenario 2, and scenario 3. With this stock availability, the company has a potential to
increase their sales when they receive a sudden demand from the buyers in that year.
However, MP1, MP8 and MP9 mine inventory always increases. The company is
suggested to focus their sales on these three qualities to avoid losses resulted from coal
combustion due to piled up inventory.
4. The relationship that can be seen from all the calculations carried out is the relationship
between the accuracy and the behavior equation of each simulation scenario. Components
with higher accuracy tend to have the same behavior in each scenario, compared to
models with low accuracy. There are significant differences in behavior on several
components, especially the MP6 Train 3, MP10 Train 3, MP8 Train 2, MP8 Train 1, MP9
Train 2, MP 11 Train 3, MP6 Train 2, MP2 Train 2, and MP9 Train 1, which are
interpreted due to the low level of accuracy in these components. It is also noted that the
components with lower accuracy tend to be the middle priority component of coal supply
to market brands.
84
|
Colorado School of Mines
|
ABSTRACT
Emergency situations, such as mine rescue operations, would benefit significantly
from the use of teleoperated vehicles. However, in an underground mine the envi-
ronment is not well suited for conventional teleoperation because the high frequency
wireless signals required for real-time video are strongly attenuated. One solution is
to provide a highly redundant network backbone, composed of many wireless nodes,
to relay information throughout the environment. Additionally, it is desirable that
such a network be flexible and mobile, so it could adapt to reach unconnected areas
and redistribute network load. In this thesis, we propose and implement a highly
capable mobile wireless network system composed of third-party Wireless-Fidelity
(Wi-Fi) mesh radios atop custom-built mobile robotic platforms. We also detail the
conversion of a Bobcat skid-steer loader into a wireless-network capable teleoper-
ated vehicle for use in an underground non-line-of-sight teleoperation scenario. The
thesis will cover the design, implementation, and evaluation of the wireless network
backbone and teleoperated remote control system for a Bobcat skid-steer loader.
iii
|
Colorado School of Mines
|
CHAPTER 1
INTRODUCTION
The field of wireless communications and teleoperated robotics has grown rapidly
in the past decade. Today, teleoperated vehicles are used for many applications
including air- and ground-based search and rescue, work in hazardous material zones,
and remote site access, among others. There has always been a strong need for
teleoperated help in disaster zones and subterranean search and rescue, because of
the inherent danger to search and rescue personal. However, the current limitations
of subterranean search and rescue robots are evident from the past disasters at the
Sago, Crandall Canyon, and Pike River mines.
1.1 Motivation
In the event of an accident, an explosion, a cave in, or whenever safety is com-
promised within a subterranean mine, rescue operations may need to occur to save
the lives of victims. However, underground rescue missions are cumbersome, time
consuming, and dangerous. Rescue workers move slowly within an underground mine
because they carry heavy equipment and must verify the safety of each section inside
the mine before they can enter, just like the scene depicted in Figure 1.1. Often,
rescue workers cannot reach victims in time and may become victims themselves.
Because of this dangerous environment and the urgency to rescue victims there
is a great need to improve the speed and safety of rescue missions. Robotic systems
and teleoperated vehicles can provide additional help in these situations and provide
a means for keeping rescue workers safe. For example, teleoperated vehicles may be
used as scouts to verify the safety of a particular hazardous area before humans enter,
to search for victims, or to reestablish communications.
1
|
Colorado School of Mines
|
loader [2] to be teleoperated over a Wi-Fi mesh network and developing a set of au-
tonomous vehicles to act as wireless relay nodes. Testing was conducted at the Edgar
experimental mine, owned by CSM.
1.4 MineSENTRY Scenario
In the MineSENTRY example scenario, an accident occurs within an underground
mineandcutsoffcommunicationandaccesstomultiplesectionsinthemine, asshown
in Figure 1.2. To accomplish a rescue effort in this scenario requires many system
components. Specifically, a teleoperated vehicle, a rescue worker base station, and
multiple robotic vehicles acting as wireless relays. Additionally, in the proposed sce-
nario the teleoperated vehicle is a Bobcat front-end loader. During the scenario,
rescue workers teleoperate a front-end loader over a mesh network to scout dangerous
areas, reestablish communications, find survivors, and possibly clear rubble. While
the teleoperated vehicle is performing its tasks, secondary autonomous vehicles wire-
lessly relay communications from the rescue worker base station to the teleoperated
vehicle and relocate themselves to maintain the best possible connection.
Figure 1.2: Example disaster scenario.
1.5 Thesis Objectives and Outline
The goals for this thesis include showing the MineSENTRY proof-of-concept for
mesh networked teleoperation in a subterranean mine and providing the documenta-
tion required so others can rebuild or extend upon the work done for this project. To
3
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.