Volcanic Earthquake Timing Using Wireless Sensor Networks

Transcription

1 Volcanic Earthquake Timing Using Wireless Sensor Networks GuojinLiu 1,2 RuiTan 2,3 RuoguZhou 2 GuoliangXing 2 Wen-Zhan Song 4 Jonathan M. Lees 5 1 Chongqing University, P.R. China 2 Michigan State University, USA 3 Advanced Digital Science Center, Illinois at Singapore 4 Georgia State University, USA 5 University of North Carolina at Chapel Hill, USA

2 Volcano Hazards Eruption in Chile, 6/4, 2011 $68 M instant damage, $2.4 B future relief. Eruptions in Iceland 2010 A week-long airspace closure [Wikipedia] 7% world population live near active volcanoes explosive eruptions/year 2

3 Volcano Monitoring Seismic activity monitoring Earthquake localization, tomography, early warning etc. Traditional seismometer Expensive (~$10K/unit), difficult to install & retrieve Only ~10 nodes installed for most threatening volcanoes! Photo credit: USGS, 3

4 Sensor Networks for Volcano Monitoring Sensor systems for volcano monitoring Harvard, Raw data & centralized analysis Short lifetime (~1 week) In-network earthquake detection [Tan 2010] Distributed seismic signal processing 83% energy reduction from raw data collection OASIS node Harvard node 4

5 Earthquake Timing Node 05 Node 04 Node 06 P-phase Time (second) Earthquake timing Source localization Seismic tomography Key to localization, seismic tomography, etc. Usually done manually, automation is expensive 5

6 Earthquake Timing Node 09 Node 10? Source localization Seismic tomography Key to localization, seismic tomography, etc. Usually done manually, automation is expensive In-situ P-phase picking w/ limited transmission Data intensive Sensors have limited compute & comm. capabilities 5

7 Seismic Signal: Sparsity 100Hz Time (second) 6

8 Seismic Signal: Sparsity 100Hz Sparsity=0.57 Time (second) wavelet K-sparse signal: s s ( k ) s 2 2 < K largest points 5% sparsity = k signal length 6

9 Seismic Signal: Sparsity 100Hz Sparsity=0.57 Time (second) Wavelet Sparsity=0.14 Time-frequency domain wavelet K-sparse signal: s s ( k ) s 2 2 < K largest points 5% sparsity = k signal length 6

10 Seismic Signal: Sparsity 100Hz Sparsity=0.57 Time (second) Wavelet Sparsity=0.14 Time-frequency domain wavelet K-sparse signal: s s ( k ) s 2 2 < K largest points 5% sparsity = k signal length Observation 1: wavelet sparsifies signal 6

11 Seismic Signal: Frequency-Time 4-level wavelet transform (length=1600) original thumbnail (length=100) Low-pass band (0, 6.25Hz) P-wave < 5Hz Time (unit: 160ms) 7

12 Seismic Signal: Frequency-Time 4-level wavelet transform (length=1600) original thumbnail (length=100) Low-pass band (0, 6.25Hz) P-wave < 5Hz Time (unit: 160ms) Original (length=1600) 7

13 Seismic Signal: Frequency-Time 4-level wavelet transform (length=1600) original thumbnail (length=100) Rough P-phase estimate Time (unit: 160ms) Original (length=1600) 7

14 Seismic Signal: Frequency-Time 4-level wavelet transform (length=1600) original thumbnail (length=100) Rough P-phase estimate Time (unit: 160ms) Original (length=1600) Observation 2: P-phase estimate from thumbnail 7

15 Seismic Signal: Diversity Node 1 sparsity=0.1 Node 10 sparsity=0.38 Earthquake 1 8

16 Seismic Signal: Diversity Node 1 sparsity=0.1 Node 10 sparsity=0.38 Earthquake 1 Node 10 sparsity=0.14 Earthquake 2 8

17 Seismic Signal: Diversity Node 1 sparsity=0.1 Node 10 sparsity=0.38 Earthquake 1 Node 10 sparsity=0.14 Earthquake 2 Observation 3: sensors have different sparsities 8

18 Outline Problem statement Approach overview Earthquake timing algorithms Performance evaluation Conclusion 9

19 Approach Overview Cluster head 10

20 Approach Overview Cluster head 10

21 Approach Overview signal sparsity preliminary pick signal sparsity preliminary pick signal sparsity preliminary pick Cluster head signal sparsity preliminary pick signal sparsity preliminary pick Lightweight signal processing algorithms Signal sparsity Preliminary P-phase from thumbnail 10

22 Approach Overview signal sparsity preliminary pick signal sparsity preliminary pick signal sparsity preliminary pick Cluster head signal sparsity preliminary pick signal sparsity preliminary pick Lightweight signal processing algorithms Signal sparsity Preliminary P-phase from thumbnail 10

23 Approach Overview Cluster head Lightweight signal processing algorithms Signal sparsity Preliminary P-phase from thumbnail Select most informative sensors to TX 10

24 Approach Overview Cluster head Lightweight signal processing algorithms Signal sparsity Preliminary P-phase from thumbnail Select most informative sensors to TX Compressive sampling & transmission 10

25 Approach Overview accurate pick Signal reconstruction accurate pick accurate pick Source localization Seismic tomography Lightweight signal processing algorithms Signal sparsity Preliminary P-phase from thumbnail Select most informative sensors to TX Compressive sampling & transmission 10

26 Outline Problem statement Approach overview Earthquake timing algorithms sensors Sensor selection & compressive sampling Performance evaluation Conclusion 11

27 Preliminary P-phase Pick 4-level wavelet transform (length=1600) thumbnail (length=100) Time (unit: 160ms) 12

28 Preliminary P-phase Pick 4-level wavelet transform (length=1600) thumbnail (length=100) preliminary pick Time (unit: 160ms) preliminary pick = 2 4 arg max p thumbnail signal energy after p signal energy before p 12

29 Preliminary P-phase Pick 4-level wavelet transform (length=1600) thumbnail (length=100) preliminary pick Map thumbnail domain back to original time domain preliminary pick = 2 Time (unit: 160ms) 4 arg max p thumbnail signal energy after p signal energy before p 12

30 Preliminary P-phase Pick 4-level wavelet transform (length=1600) thumbnail (length=100) preliminary pick Map thumbnail domain back to original time domain preliminary pick = 2 Time (unit: 160ms) 4 arg max p thumbnail signal energy after p signal energy before p Lightweight: O(signal length) Suitable for resource-constrained sensors 12

31 Outline Problem statement Approach overview Earthquake timing algorithms sensors Sensor selection & compressive sampling Performance evaluation Conclusion 13

32 Impact of Timing on Source Localization Source localization Basis for many volcano monitoring applications Complex non-linear inverse problem z z 2 1 V z 3 t t t ray tracing( z, z, V i = i 0 ) 2 1 t 3 z 0 Information-theoretic error metric ( ) ) T 1 E = tr GG scaled Fisher matrix: z i, z 0 14

33 Impact of Timing on Source Localization Source localization Basis for many volcano monitoring applications Complex non-linear inverse problem sensor position z z 2 1 V z 3 t t t ray tracing( z, z, V i = i 0 ) 2 1 t 3 z 0 Information-theoretic error metric ( ) ) T 1 E = tr GG scaled Fisher matrix: z i, z 0 14

34 Impact of Timing on Source Localization Source localization Basis for many volcano monitoring applications Complex non-linear inverse problem sensor position source location z z 2 1 V z 3 t t t ray tracing( z, z, V i = i 0 ) 2 1 t 3 z 0 Information-theoretic error metric ( ) ) T 1 E = tr GG scaled Fisher matrix: z i, z 0 14

35 Impact of Timing on Source Localization Source localization Basis for many volcano monitoring applications Complex non-linear inverse problem sensor position source location z z 2 1 V z 3 t t t ray tracing( z, z, V i = i 0 ) 2 1 t 3 z 0 Information-theoretic error metric volcano model ( ) ) T 1 E = tr GG scaled Fisher matrix: z i, z 0 14

36 Dynamic Sensor Selection Find a subset of sensors Sto minimize Es.t. i S c i m(sparsity of sensor i) C 15

37 Dynamic Sensor Selection Find a subset of sensors Sto minimize Es.t. i S c i m(sparsity of sensor i) C unit TX cost 15

38 Dynamic Sensor Selection Find a subset of sensors Sto minimize Es.t. i S c i m(sparsity of sensor i) C unit TX cost TX volume 15

39 Dynamic Sensor Selection Find a subset of sensors Sto minimize Es.t. i S c i m(sparsity of sensor i) C unit TX cost TX volume cost budget 15

40 Dynamic Sensor Selection Find a subset of sensors Sto minimize Es.t. i S c i m(sparsity of sensor i) C unit TX cost TX volume cost budget Brutal-force search 8 seconds on Imote2 for 16 sensors Information gain diminishes for larger clusters 15

41 Compressive Sampling (CS) n m x n = m random matrix original compressed Apply CS to wavelet coefficients Known TX volume before compression m = 1.5 sparsity n Unselected sensors avoid compression overhead 16

42 Compressive Sampling (CS) n m x n = m random matrix original compressed Apply CS to wavelet coefficients Known TX volume before compression m = 1.5 sparsity n best trade-off b/w TX volume and signal reconstruction error Unselected sensors avoid compression overhead 16

43 Outline Problem statement Approach overview Earthquake timing algorithms Performance evaluation Testbedexperiments Extensive trace-driven simulations Conclusion 17

44 Testbed Experiments Implementation on 12 TelosB Seismic data from Mt St Helens -> mote flash Real-time data 100 Hz 4 Execution time (second) End-to-end delay < 3 seconds Sensor ID 18

45 Error metric Trace-driven Simulation Data traces from 12 sensors on Mt St Helens 30 significant earthquakes in 5.5 months Lance [SenSys 08] our sensor selection approach TX bound (# of pkts) TX bound (# of pkts) # of selected senso ors Configurable trade-off between system performance and energy consumption 19

46 Impact of Packet Loss = x Relative reconstruction erro or (%) compressed signal Compressive Sensing Lossy compression: encodes largest wavelet coefficients Packet reception ratio (%) reconstructed signal CS is resilient to packet loss! 20

47 Impact of Packet Loss received = x Relative reconstruction erro or (%) compressed signal Compressive Sensing Lossy compression: encodes largest wavelet coefficients Packet reception ratio (%) reconstructed signal CS is resilient to packet loss! 20

48 Impact of Packet Loss received = x Relative reconstruction erro or (%) compressed signal Compressive Sensing Lossy compression: encodes largest wavelet coefficients Packet reception ratio (%) reconstructed signal CS is resilient to packet loss! 20

49 Accuracy of Timing fine-grained pick on original fine-grained pick on reconstructed 21

50 Accuracy of Timing fine-grained pick on original fine-grained pick on reconstructed 16% data TX 0.6 km localization error 21

51 Conclusions Energy-efficient earthquake timing Lightweight algorithms for sensors Dynamic sensor selection Compressive sampling Testbed experiments Feasibility of our approach on motes Trace-driven simulations Accurate timing with 16% data transmitted 22

52 Hierarchical Network Architecture sensor coordinator cluster 6.7km se ensor / coordinator STA/LTA detector Bayesian detector 500 nodes on Tungurahua, Ecuador, 2015 [VolcanoSRI project] Sensors Limited capability, large spatial coverage Coordinators Powerful, limited number # earthquakes per day MCU & radio energy ratio TelosB vs. Imote2 23

53 Earthquake Source Localization % of tra ansmitted data Source loc calization error (km) Packet reception ratio 85% # of packets # of packets 0.2 Source localization result for an earthquake 16:56:47 Nov 03 Mt St Helens 24

54 Earthquake Source Localization % of tra ansmitted data Source loc calization error (km) Packet reception ratio 85% # of packets # of packets 0.2 Source localization result for an earthquake 16:56:47 Nov 03 Mt St Helens Localization error below 1km, common in volcano seismology Only 16% data transmission 24