Entry Name: SMU-Team JKY-MC2
VAST Challenge 2017
Mini-Challenge 2
Team Members:
Dr. Kam Tin Seong, Singapore
Management University, tskam@smu.edu.sg
Kishan Bharadwaj Shridhar, Singapore Management University,
kishanbs.2016@mitb.smu.edu.sg
Ong Guan Jie Jason, Singapore Management University, jason.ong.2016@mitb.smu.edu.sg
Zhang Yanrong, Singapore Management University, yrzhang.2016@mitb.smu.edu.sg , PRIMARY
Student Team: YES
Tools Used:
· Tableau
· Excel
Approximately how many hours were spent working on this submission
in total?
200
May we post your submission in the Visual Analytics Benchmark
Repository after VAST Challenge 2017 is complete? YES
Tableau Workbook:
Questions
MC2.1 – Characterize the sensors’ performance and
operation. Are they all working properly at
all times? Can you detect any unexpected behaviors of the sensors through analyzing the readings they capture? Limit your
response to no more than 9 images and 1000 words.
Monitors’ Performance
There are a total of 9 monitors
installed in the vicinity of the four factories. The trellis plot allows us to
observe the performance of the monitors at a high level.
Fig 1.1 shows a trellis plot capturing emission readings of the
chemical AGOC-3A for the month of April. Immediately we can deduce that
monitors 1 and 2 are working normally because emission readings are very close
to the average line. Similarly, monitors 4 - 6 should be working as intended
because fluctuations about the average line seem normal.
Fig 1.1
We can verify the findings above by performing a drill-in analysis
of the trellis plot which brings us to the cycle plot illustrated in Fig 1.2.
We can see that emission readings are relatively flat for monitors 1 and 2
whereas fluctuations are regular for monitors 4 - 6 which lends further
credence to the earlier findings.
Fig 1.2
Unexpected Behavior
Scenarios whereby monitors are exhibiting unexpected behavior could be instances of sudden spikes in
readings happening erratically during the month. Fig 1.3 shows an illustration
of this. The trellis plot above shows that there is a sudden spike in emission
readings for chemical AGOC-3A on day 17 for the month of april. The cycle plot below allows a deeper understanding
of what is happening and we can see that there were two occurrences of
abnormally high readings captured at 06:00 and 07:00 on the 17th April.
Fig 1.3
Another clearer example can be seen in Fig 1.4. We can see that
there is only one instance of abnormally high reading for chemical Methylosmolene captured by both graphs on the 11th of
April at 03:00. This is unexpected because the graph depicts a relatively
smooth trend for the rest of the month.
Fig 1.4
However, there will also be blurred scenarios and are not so
easily to discern unlike the above examples. For example. Fig 1.5 shows that
emission readings for chemical AGOC-A are relatively flat and stable for
monitors 1,2 and 3. However for each of them, there is one occurrence of
abnormally high reading detected. Is this a simple case of faulty monitors? The
cycle plot suggests otherwise. This anomalous event actually
occurred at the same timing, 06:00 on 5th December for all
monitors. Is it possible that all 3 monitors who are within close proximity to each other were faulty at the same
time or is there some other underlying reason?
Fig 1.5
MC2.2 – Now turn your attention to the chemicals
themselves. Which chemicals are being detected by the sensor
group? What patterns of chemical releases do you see, as being
reported in the data?
Limit your response to no more than 6 images and 500 words.
Detected Chemicals
We can leverage on the trellis plot to have a quick overview which
chemicals are being detected by which monitors. Fig 1.6 shows a representation
of the detection of chemical AGOC-3A across the 3 months.
Fig 1.6
Monitors 3-7 and 9 are able to detect
AGOC-3A fairly well. Monitors 3 and 6 were picking up a highest concentration
for April at an average of 32.8ppm daily. Monitor 3 continued to pick up the
highest concentration at 51.6ppm for August but monitor 4 picked up a
higher concetration at 60.3ppm for
December.
Similarly for Appluimonia, monitors 3 and 4 were picking up a higher
concentration of this chemical compared to the other monitors across these 3
months. Monitors 5-9 were able to detect a
considerably but lesser concentration as well.
For Chlorodinine, all monitors
except for 1 were able to detect a
considerable amount of this chemical. Monitor 3 continued to pick up the
highest concentration for all months.
Monitors 3,6 and 7 were able to detect Methylosmolene at a much higher concentration compared
to others for April. But at the later months, monitors 4 and 5 were starting to
pick up a substantial amount of this chemical as well with the former detecting
the highest concentration out of all monitors during December.
Release Patterns
The cycle plot and horizon graph are used to observe the time
patterns in the release of the different chemicals.
· AGOC-3A
This chemical is released typically between 06:00 to 22:00 based
on the 3 months of data. Also, the level of activity is higher between the 5th
to 22nd of each month.
Fig 1.7
· Appluimonia
This chemical is being released constantly throughout the month
and exhibits the same trend for all 3 months. Moreover, the release of this
chemical has increased dramatically during the month of December with almost
all the monitors registering an increase compared to the previous two months.
· Chlorodinine
Similar to the
previous chemical, Chlorodinine is also
being released constantly throughout the 3 months of data. The release of this
chemical typically spikes near the start and approaching the end of the month.
Also, the level of concentration detected by monitor 4 has increased
dramatically from April to December.
Fig 1.8
· Methylosmolene
This chemical's release pattern is typically between 22:00 and
06:00 based on the 3 months of data. This is coincidentally the opposite
pattern of AGOC-3A which suggests that both chemicals may share a relationship
with one another.
Fig 1.9
MC2.3 – Which factories are responsible for which chemical
releases? Carefully describe how you determined this using all the data you
have available. For the factories you identified, describe any observed
patterns of operation revealed in the data.
Limit your response to no more than 8 images and 1000 words.
Factories Responsible for Chemical Releases
The way the dashboard can be used to trace the release of the
pollutants to the polluters is to look for spikes in the monitors readings via
the cycle plot then set the filters for the air plume model to the
corresponding day and hour. After that, we observe if the polygons contain any
factories in their trajectories. This could be a possible indication that the
factory is responsible for releasing that pollutant. The tip of the polygon
represents the direction the wind is coming from and the flat edge represents
the stream of wind detected by the monitor. The angle parameter is used to
adjust the spread of the wind in certain ambiguous situations which will be
further elaborated in the later part. By default, the spread is assumed to be
10 degrees. The map used for the air plume model is shown below.
· AGOC-3A
Roadrunner and Kasio were
located very close to one another which makes it very difficult to discern
whether there is only one actual factory releasing the chemical or both are
releasing the chemicals. The dashboard has managed to pinpoint the potential
culprits to both of these companies multiple
times but due to their close proximity which is highlighted in Fig 2.1, it
makes it very difficult to find cases to isolate one factory. On 6th April at
06:00, the wind was blowing from 270 degrees with a speed of 0.9m/s and monitor
6 captured a reading of 228.8ppm. However, the polygon we are interested in
which is colored in yellow overlaps two
factories in its trajectory therefore we are able to tell
if one or both contributed to the pollution.
Fig 2.1
There is one particular instance illustrated
in Fig 2.2 which manages to provide considerable evidence to lay charges on one
of the factories. On the 22nd April at 09:00, the polygon overlaps only one
factory in its trajectory which is subsequently captured by monitor 9.
Therefore, we can be fairly positive Roadrunner
is one of the factories releasing AGOC-3A.
Fig 2.2
· Appluimonia
One of
the highest level of concentration was captured on 29th April at 09:00. The
reading was 26.85ppm which is highest for that month and wind speed was 0.2m/s.
This suggests that Radiance is responsible for emitting this chemical since it
is the only factory covered by the trajectory.
Fig 2.3
However,
at this particular point of time, the
trajectory of the polgyon is very close to
both Radiance and Indigo. In fact, if we were to adjust the angle parameter to
make the polygon has a larger spread, the two polygons will overlap and
moreover owing to the low wind speed, it is very possible that chemicals
emitted by Indigo may be carried by the polygon 6 hence Radiance is innocent.
Fig 2.4
On the
13th August at 09:00, the reading was 15.29ppm and the polygon contained only
Indigo in its trajectory hence this provides ample evidence that Indigo is one
of the factories responsible for producing this chemical.
Fig 2.5
· Chlorodinine
One of
the higher readings fell on the 4th April at 12:00 am. The
yellow polygon contains only roadrunner in its trajectory so this suggested
that the company is responsible for producing Chlorodinine.
Fig 2.6
· Methylosmolene
Tracing this pollutant back to its polluter faces a similar
problem as that of AGOC-3A because Roadrunner and Kasios are
situated too closely. Fig 2.7 gives explicit evidence that Kasios is the factory emitting this chemical based on
readings captured by monitor 6 on the 25th April at 03:00.
Fig 2.7
Observed
Patterns of Operations
Using the
chemical emission level as a proxy for factory operations, Roadrunner seems to
be operating more actively between 06:00 to 18:00 during April.
Kasios on the other hand appears to also be
operating more actively during the wee hours of the morning typically between
0:00 to 06:00 and this pattern is consistent for all 3 months.