Entry Name:  "Purdue-Tang-MC2"

VAST Challenge 2017

Mini-Challenge 2

Team Members:

Hui Tang, Purdue University, West Lafayette, USA, tang227@purdue.edu                PRIMARY

Wenjie Wu, Purdue University, West Lafayette, USA, wu1116@purdue.edu 

Zheng Zhou, Purdue University, West Lafayette, USA, zhou85@purdue.edu 

Sijin Wang, Purdue University, West Lafayette, USA, wang2283@purdue.edu 

Andrew Aijun Huang, Purdue University, West Lafayette, USA, huan1004@purdue.edu 

Yafeng Niu, Southeast University, China, niu29@purdue.edu 

Yingjie Chen, Purdue University, West Lafayette, USA, victorchen@purdue.edu 

Zhenyu Qian, Purdue University, West Lafayette, USA, qianz@purdue.edu 

Student Team:  YES

Tools Used: VanillaJS, D3.js, PHP, MySQL, Microsoft Excel

Approximately how many hours were spent working on this submission in total?

200 hours

May we post your submission in the Visual Analytics Benchmark Repository after VAST Challenge 2017 is complete?

YES

Video

https://goo.gl/ufGkU2 

Live Demo

https://goo.gl/jU1AfB - WindNebula: Vectorial-Temporal Analysis for Environmental Assessment

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.

1. In general, sensors that are closer to the factory tend to have larger readings more frequently. Sensor readings were highly related to wind directions (Fig 2.1.1). Not all sensors operated normally all the time as some showed unique patterns while others did not show the similar patterns.

Fig 2.1.1 The vectorial and temporal view of the WindNebula system. Sensor reading dots are placed inside the white circles according to their readings and corresponding wind directions at the moment (Upper part). Temporal view of readings are juxtaposed by sensors for sensor-wise comparison (Bottom part).

2. Compared to other sensors, sensor 1 showed relative low readings for four chemicals all the times (Fig 2.1.2). The major cause is its geographical location. First, sensor 1 is far from the factories and thus the pollution condition at that distance was relatively low. Second, during the given three months, wind towards northeast direction is way more frequent (according to the wind frequency heat map) and sensor 1 is located at the far west part of the area.

Fig 2.1.2 The detailed temporal view of the sensor 1 by month and corresponding wind frequency heat maps on the top. With a remote location and calm meteorological condition, sensor 1 detected low readings at most of the time.

3. Possible sensor malfunction of sensor 4. Sensor 4 had a shifting baseline of readings over time (Figure 2.1.3). The readings were having increasing range throughout April, August, and December for all the four chemicals. We assume that the readings had been increasing continuously from April to December. Other sensors did not show sucuh a trend.

Fig 2.1.3 The detailed temporal view of the sensor 4 by month and a side-by-side temporal plot with all sensors. Sensor 4 reported a pattern of shifting baseline while others did not.

4. Possible sensor malfunction of sensor 5. The readings were magnifying throughout April, August, and December evenly and continuously (Figure 2.1.4). This increasing trend were more subtle compared to sensor 4. No other sensor reported this trend. Sensor 9 also reported a magnifying condition but not as even as sensor 5’s change.

Fig 2.1.4 The detailed temporal view of the sensor 5 by month and a side-by-side temporal plot with all sensors. Sensor 5 reported a pattern of magnifying range over months while others did not.

5. Compared with other sensors, sensor 3 and 6 had much less near-zero readings (Fig 2.1.5). Especially when we compare with those sensors (5, 7, and 9) that had similar geographical locations, sensor 3 and 6’s sparse reading gaps are pretty strange. We suspect that was also a malfunction (such as abnormal high threshold of sensitivity).

Fig 2.1.5 The side-by-side temporal plot of all sensors in logarithm scale. Sensor 3 and 6 reported sparse readings at low values while others did not.

6. Possible power failure event. At midnight 00:00 on Apr 3, Apr 7, Aug 3, Aug 5, Aug 8, Dec 3, and Dec 8, all sensors for all four chemicals had no readings (Fig 2.1.6). We assume there was a power failure event for the sensors during those midenight moments. Figure 2.1.6 shows a sensor 2 as an example.

Fig 2.1.6 The detailed temporal view of the sensor 2 by month. Moments of missing reading are marked at the bottom time line as hollowed dots. When all four chemicals were missing readings at the same time, we consider a power failure event took place at this moment. We use vertical lines to highlight such moments.

7. Quantity of larger readings for sensor 6 did not observe the dispersion ‘monotonically decreasing’ rule. There was a gap between normal readings and large readings (Fig 2.1.7).

Fig 2.1.7 The side-by-side temporal plot of all sensors. Sensor 6 did not show the pattern of sparse gradient from low readings to high readings.

8. Sensor 3 had erroneous readings on Appluimonia and Chlorodinine (Fig 2.1.8). The reason why these two chemicals anomalies cannot be considered as ‘background noise’ is because no other sensor detected this obvious ‘background’ readings.

Fig 2.1.8 The detailed temporal views of the sensor 3 by month and by chemical. Appluimonia and Chlorodinine views indicate that readings are detected in many random directions, and not correspond to the directions from the four factories. But the AGOC-3A and Methylosmolene views does not show such a pattern.

9. Sensor 9 had sudden increased range of readings for all four chemicals from August 23 8:00 onwards. This increment carried out throughout December (Fig 2.1.9) . No other sensors showed this trend. We need more evidence to form a reasonable hypothesis.

Fig 2.1.9 The detailed temporal view of the sensor 9 by month and a side-by-side temporal plot with all sensors. Each sensor had its own stable baseline, but sensor 9 showed a sudden raised baseline from August 23 8:00.

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.

1. All chemicals had been detected by all sensors. Figure 2.2.1 verified this conclusion.

Fig 2.2.1 The image above shows side-by-side temporal plots with all sensors for all four chemicals. All the four chemicals including AGOC-3A, Appluimonia, Chlorodinine and Methylosmolene had been detected by all the sensors.

2. Contradicting AGOC-3A readings in pairs occurred at times on all nine sensors. Mostly during daytime between 6:00 to 22:00. Meanwhile, missing reading of Methylosmolene for the exact same moment were found. We believe some of the reading of Methylosmolene were misreported as AGOC-3A. This could be the similar characteristic within the same chemical family according to the chemical description. The sensors may had confused and misreported. Figure 2.2.2 uses sensor 6 and sensor 9 as examples to visualize this situation.

Fig 2.2.2 The detailed temporal view of the sensor 6 and sensor 9. Misreported data are labeled as AGOC-3A poles (orange) connecting two contradictory readings at the same time. Corresponding hollowed cyan Methylosmolene circle marked missing data at the bottom timeline. 

3. Large Methylosmolene readings occurred only during 22:00 and 5:00 on all nine sensors. An example from sensor 6 demonstrates this pattern (Figure 2.2.3). However, taking into account of the misreport situation mentioned above (Question 2 #2), this ‘night-only’ pattern might not be true to present Methylosmolene concentration.

Fig 2.2.3 The detailed temporal view of the sensor 6 for AGOC-3A and Methylosmolene. The ‘Day-only’ for AGOC-3A and ‘Night-only’ for Methylosmolene pattern could be cancelled out if we consider the misreport situation.

4. Sensor 5 and 6 are the only sensor group that can activily detect the change of AGOC-3A. Compared with them, other sensors were not sensitive to report the changes. Figure 2.2.4 uses sensor 7 as an example to demonstrate this situation.

Fig 2.2.4 The detailed temporal view of the sensor 7 for AGOC-3A. Besides the misreported AGOC-3A, there were actually no large AGOC-3A detected.

5. Because of their geographical locations (distance from the factories and wind directions), sensor 1 and 2 failed to detect any significant changes of all chemicals. Although they were able to detect small values of readings, but during the three months, they can’t report any useful findings.

Fig 2.2.5 The detailed temporal view of the sensor 1 and sensor 2. Besides the small readings, there were actually no large meaningful readings detected.

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.

1. We were able to identify the a major and a minor kinds of chemicals emitted by each factory according to the downwind direction between the factories and the sensors. The sensor reading should have the most concentration along that direction. By connecting a line from the factory to the sensors, if the high readings from most of the sensors were aligned with those lines, then we can conclude that factory was responsible for that chemical. Chemicals releasing ranked by concentration for each factory are listed as follows:

• Roadrunner: Chlorodinine (Fig 2.3.1.1 Top), minor trace of AGOC-3A (Fig 2.3.1.1 Bottom)
• Kasios: AGOC-3A (Fig 2.3.1.2 Top), minor trace of Methylosmolene (Fig 2.3.1.2 Bottom)
• Radiance: AGOC-3A (Fig 2.3.1.3 Top), minor trace of Appluimonia (Fig 2.3.1.3 Bottom)
• Indigo: Appluimonia (Fig 2.3.1.4 Top), minor trace of Methylosmolene (Fig 2.3.1.4 Bottom)

Fig 2.3.1.1 Top: Vectorial view of the chemical Chlorodinine together with connection lines between Roadrunner fitness electronics factory and all sensors. Emission of Chlorodinine from Roadrunner was confirmed because multiple large readings were found along that direction. Bottom: Minor emission of AGOC-3A from Roadrunner was found by the same approach.

In terms of operation pattern of Roadrunner, we chose the most representative sensor 6 of Chlorodinine releasing temporal plot to show the pattern (Fig 2.3.1.1.a).

Fig 2.3.1.1.a The detailed temporal view of sensor 6 of Chlorodinine indicates most large readings occurred during weekdays.

Fig 2.3.1.2 Top: Vectorial view of the chemical AGOC-3A together with connection lines between Kasios office furniture factory and all sensors. Emission of AGOC-3A from Kasios was confirmed because multiple large readings were found along that direction. Bottom: Minor emission of Methylosmolene from Kasios was found by the same approach.

In terms of operation pattern of Kasios, we chose the most representative sensor 3 of AGOC-3A releasing temporal plot to show the pattern (Fig 2.3.1.2.a).

Fig 2.3.1.2.a The detailed temporal view of sensor 3 of AGOC-3A shows no obvious operating pattern. The factory were seemingly operating all week along.

Fig 2.3.1.3 Top: Vectorial view of the chemical AGOC-3A together with connection lines between Radiance ColourTek factory and all sensors. Emission of AGOC-3A from Radiance was confirmed because multiple large readings were found along that direction. Bottom: Minor emission of Appluimonia from Radiance was found by the same approach.

In terms of operation pattern of Radiance, we chose the most representative sensor 8 of AGOC-3A releasing temporal plot to show the pattern (Fig 2.3.1.3.a).

Fig 2.3.1.3.a The detailed sensor 8 temporal view of AGOC-3A shows no obvious operating pattern. The factory were seemingly operating all week along.

Fig 2.3.1.4 Top: Vectorial view of the chemical Appluimonia together with connection lines between Indigo Sol Boards factory and all sensors. Emission of Appluimonia from Indigo was confirmed because multiple large readings were found along that direction. Bottom: Minor emission of Methylosmolene from Indigo was found by the same approach.

In terms of operation pattern of Indigo, we chose the most representative sensor 9 of Appluimonia releasing temporal plot to show the pattern (Fig 2.3.1.4.a).

Fig 2.3.1.4.a The detailed temporal view of sensor 8 of AGOC-3A shows no obvious operating pattern. The factory were seemingly operating all week along.

2. Monthly patterns of operation were not clear. The following image shows the average readings during daytime/nighttime and weekday/weekend. We may develop a false sense that certain AGOC-3A releasing factories (Kasios and Radiance) operated more on daytime than nighttime while some Methylosmolene releasing factories (Kasios and Indigo) operated more during the night time. Nevertheless, considering the misreporting condition (Question 2 #2), this hypothesis may not be true.

Fig 2.3.2 The average readings indicator for each individual sensor for each single chemical for weekday/weekend and day/night summarize the chemical releasing pattern.