Monitoring Volcanic Eruptions with a Wireless Sensor Networks Geoffrey Werner-Allen, Jeff Johnson,...

Post on 11-Jan-2016

212 views 0 download

Transcript of Monitoring Volcanic Eruptions with a Wireless Sensor Networks Geoffrey Werner-Allen, Jeff Johnson,...

Monitoring Volcanic Eruptions with a Wireless Sensor Networks

Geoffrey Werner-Allen, Jeff Johnson, Mario Ruiz, Jonathan Lees, and Matt Welsh

Harvard UniversityEWSN’05

Presented by Tim

Outline Introduction Background System Design Deployment Distributed Event Detection Evaluation Conclusion

Introduction Volcanic monitoring has a wide range of goals,

related to both scientific studies and hazard monitoring.

Volcanologists currently use wired arrays of sensors to monitor volcanic eruptions.

Wireless sensor networks have the potential to greatly benefit studies of volcanic activity.

Background Infrasound (Infrasonic wave)

Sound with very low frequency (1~50Hz) Very high amplitude but not audible

Seismic wave Wave travels through the Earth, often as the

result of an earthquake or explosion

Volcanic Monitoring

Challenges and Issues Existing data loggers store data locally

e.g., 1 or 2 Gb microdrives, store about 15 days' worth of data

Must trek up to the station to retrieve the data Usually very inaccessible: can take several hours to drive/hike in

Very high power consumption Two car batteries plus solar panels to recharge

Very expensive Individual data logger costs thousands of $$$

Still need PCs/laptops to process and store data permanently

Hard to deploy large number of stations Size, cost, power requirements,...

Opportunities for wireless sensor networks Data sampling rates of ~100 Hz Very small, low power, easy to deploy Can put out a larger number of sensors in an a

rea Can customize software on the motes for capt

ure, preprocessing, etc.

Outline Introduction Background System Design Deployment Distributed Event Detection Evaluation Conclusion

System Architecture

Infrasound Node Sample data continuously at

102.4Hz A set of 25 consecutive sampl

es is packed into a 32-byte packet and transmitted at approximately 4 Hz.

The aggregator will send acknowledgement back. If source node does not receive ack, it’ll retransmit up to 5 times.

Aggregator Node

GPS Receiver Node Motes record sample # and GPS time seq # in

message Can be used to align samples from each mote

Time Regression Uncertainties

The sampling rate of individual note may vary slightly over time, due to changes in temperature and battery voltage.

The log do not record the precise time. Apply a linear regression to the data log

stream and map individual sample to a “true ” time.

Physical Packaging

Outline Introduction Background System DesignDeployment Distributed Event Detection Evaluation Conclusion

Volcano Tungurahua Active volcano in central

Ecuador – 5018 m Site of much ongoing se

ismological research

Deployment Three infrasound nodes, one central aggregator node

and a GPS receiver. The GPS receiver and FreeWave modem were powere

d by a 12 V car battery. All other nodes were powered by 2 AA batteries.

The distance between sensors and observatory is about 9km.

The deployment was active from July 20–22, 2004 and collected over 54 hours of infrasonic signals.

Deployment

Deployment

Deployment

Data Analysis- Loss Rate

• Weather conditions (e.g., rain) affected radio transmission.• Mote 4 experienced very low loss, due to its position with line-of-sight to the receiver.• Mote 3 experienced higher loss, probably due to antenna orientation.

Data Analysis- Correlation

The result of wireless sensor array shows high correlation with wired station.

Outline Introduction Background System Design DeploymentDistributed Event Detection Evaluation Conclusion

Distributed Event Detection The initial deployment is not feasible for

larger arrays deployed over long period of time.

To save bandwidth and energy, it is desired to avoid transmitting signals when the volcano is quiescent.

Mechanism Each node samples data continuously at 102.4

Hz. When the local event detector triggers, the node

broadcasts a vote message. If any node receives enough votes from its

neighbor nodes, it initiates global data collection by flooding a message to all nodes in the network.

Token-based scheme for scheduling transmissions. The order depends on node ID.

Local Detector Design Threshold-based detector Exponentially weighted moving average

based detector

Local Detector Design Threshold-based detector

Triggered whenever a signal rises above Thi and falls below another Tlo during some time window W.

Because it relies on absolute thresholds, it is sensitive to particular microphone gain on each node.

Local Detector Design Exponentially weighted moving average based detector

For each sample, calculate two moving averages with different gain parameters, αshort ,αlong ,and compare the ratio of the two averages.

e.g., (αshort = 0.05,αlong =0.002) If the ratio exceeds some threshold T (i.e., the short-term

average exceeds the long-term average by a significant amount), the detector is triggered.

averagesampleaverage )1(

Outline Introduction Background System Design Deployment Distributed Event DetectionEvaluation Conclusion

Evaluation Use 8 mica2 nodes in the lab, but only 4

nodes with infrasound sensor board. The infrasound signals were produced by

closing the lab door. Three parts

Energy usage Bandwidth usage Detector accuracy

Energy usage

Each node exhibits a baseline current draw of about 18mA and supply voltage is 3 V.

Assuming that nodes detect a correlated signal every ½ hours, and locally vote at twice this rate.

mWnPPPd

mWPPc

txsendtxvote

txtx

0744.54)1800/16.2(60900/16.2183183

64.6216.24183183

Bandwidth usage Continuous sampling sche

me consumes nx4x32 bytes/sec of bandwidth (n:# of nodes, each node tra

nsmit one pkt every ¼ sec, size of pkt :32bytes)

Because of the low frequency of eruptions, distributed event detection uses less bandwidth.

Detector Accuracy Fed the detectors with the complete trace of d

ata recorded on Tungurahua.

Future Work & Conclusion Seismology presents many exciting opportunit

ies for wireless sensor networks. To expand the number of sensors in the array

and distribute them over a wider aperture. The long-term plans are to provide a permane

nt, reprogrammable sensor array on Tungurahua.

My Comments The idea is simple but it’s hard work to

deploy the motes in such a place. To do research needs lots of passion.

The first mote-based application to volcanic monitoring! Provide a wealth of experience to develop

more sophisticated tools.