Distributed Structural Health Monitoring A Cyber Physical System Approach

Transcription

1 Distributed Structural Health Monitoring A Cyber Physical System Approach Chenyang Lu Department of Computer Science and Engineering

2 American Society for Civil Engineers 2009 Report Card for America's Infrastructure Bridges C Dams D Levees D Rail C Roads D More than 26%, or one in four, of the na>on's bridges are either structurally deficient or func>onally obsolete. America's Infrastructure GPA: D Es>mated 5 Year Investment Need: $2.2 Trillion 2

3 Structural Health Monitoring (SHM) Detect and localize damages to structures Wireless sensor networks can monitor at high temporal and spa>al granulari>es Key Challenges Computa>onally intensive Resource constraints Long term monitoring 3

4 Related Work Wisden [Xu SenSys'04] Services for reliable transmission of sensor data Golden Gate Bridge [Kim IPSN'07] 46 hop network deployed along Golden Gate Bridge BriMon [Chebrolu MobiSys'08] Trains as data mules Torre Aquila Deployment [Cerio` IPSN'09]] Monitoring heritage buildings 4

5 Existing Approaches Centralized approach: stream all sensor data to base sta>on for processing. Useful for model valida>on. But too energy consuming for long term monitoring? Example: Golden Gate Bridge project. Nearly 1 day to collect enough data for one computa>on Life>me of 10 weeks w/4 x 6V lantern baiery Separate designs of sensor networks and SHM algorithm. Primarily focus on data transport issues. Not concerned with SHM algorithm used for damage localiza>on. 5

6 Distributed Architecture Dilemma Too much sensor data to stream to the base sta>on SHM algorithms are too complex to run en>rely on sensors Raw Data Distributed Architecture Performs part of computa>on on sensor nodes Send (smaller) intermediate results to base sta>on Base sta>on completes computa>on Par>al Results 6

7 Cyber Physical Co design The term cyber physical systems refers to the >ght conjoining of and coordina>on between computa>onal and physical resources. [NSF] Raw Data Cyber Physical Co design of SHM Systems Iden>fy an SHM approach that allows efficient distributed implementa>on over sensor networks. Op>mally map SHM algorithms onto distributed sensor network architecture. Par>al Results 7

8 SHM Algorithm Damage Localization Assurance Criterion (DLAC) Damage Localiza>on [Messina 96] Iden>fy structure s natural frequencies based on vibra>on data. Signature of structure s health Match natural frequencies to structural models with damages of interest. Allow op>mal par>>oning of computa>on stages between sensors and the base sta>on. Minimize energy consump>on Subject to resource constraints 8

9 Amplitude 0!500!1000!1500! Time(s) Vibration Data Input DLAC Algorithm 9

10 Amplitude(dB) ! Frequency (Hz) FFT + Power Spectrum Analysis DLAC Algorithm 10

11 "%! "#! )*+,-./0,12-34 "!! (! '! %! #!!!#!.! "! #! $! %! &! 6-,73,819.:;<= Curve Fitting DLAC Algorithm 11

12 DLAC A mathema>cal model of the structure is created offline Used to predict effect of structural damage on natural frequencies Natural frequency data input: Healthy structure Healthy model Model damaged at different discrete loca>ons Damaged structure 12

13 Highest correla.on to damage at Loca.on Element Position DLAC Output DLAC Algorithm 13

14 Distributed Architecture 1. Performs part of computa>on on sensors 2. Send intermediate results to base sta>on 3. Base sta>on completes computa>on Raw Data Where should the algorithm be par22oned? Partial Results 14

15 (3a) Coefficient Extrac>on (3b) Equa>on Solving D Integers 5*P Floats Healthy Model (1) FFT (2) Power Spectrum (3) Curve Fi`ng (4) DLAC 2D Floats D Floats P Floats D: # of samples P: # of natural freq. (D» P) Damaged Location Data Flow Analysis DLAC Algorithm 15

16 4096 bytes (1) FFT D: 2048 P: 5 (3a) Coefficient Extrac>on 100 bytes (3b) Equa>on Solving (2) Power Spectrum Effec>ve compression ra>o of 204:1 (3) Curve Fi`ng 8192 bytes 4096 bytes 20 bytes Integer: 2 bytes Float: 4 bytes Healthy Model (4) DLAC Damaged Location Data Flow Analysis DLAC Algorithm 16

17 Implementation Sensor plavorm: Intel/Crossbow Imote2 + ITS400 sensor board MHz XScale CPU 32 MB ROM, 32 MB SDRAM CC compliant radio 3 axis accelerometer on sensor board Data collec>on and processing applica>on wriien with TinyOS KB ROM, 71 KB RAM 17

18 Evaluation: Truss 5.6m steel truss structure at UIUC m long bays, on 4 rigid supports 11 Imote2s aiached to frontal pane 1 DLAC WS #32 1 DLAC WS #45 1 DLAC WS #67 1 DLAC WS #28 1 DLAC WS #35 1 DLAC WS # X = 3 Y = X = 3 Y = X = 3 Y = X = 3 Y = X = 3 Y = X = 3 Y = Damage correctly localized to 0.1 third 0.1 bay 0.1 Truss Frontal Panel Wireless Sensor Truss Central Bay Position Truss Central Bay Position Truss Central Bay Position Truss Central Bay Position Truss Central Bay Position Truss Central Bay Position 18

19 Centralized Decentralized Sampling Computa>on Communica>on Energy consump.on (mah) Energy Consumption Evaluation 19

20 Equa>on Solving Coefficient Extrac>on Sampling Computa>on Communica>on Power Spectrum FFT Raw Data Collec>on Energy Consump.on (mah) Energy Consumption Evaluation 20

21 Centralized Decentralized Sampling Computa>on Communica>on Latency (ms) Latency Evaluation 21

22 ROM RAM Raw Data Collec>on FFT Power Spectrum Coefficient Extrac>on Equa>on Solving Size (bytes) Memory Consumption Evalua>on 22

23 Summary Cyber physical co design of a distributed SHM system. Reduces energy consump>on by 71% Reduces latency by 66% Implemented on imote2 using <1% of its memory Effec>vely localized damage on two physical structures. Demonstrated the promise of cyber physical co design of sensor network systems. 23

24 Acknowledgement Computer Science: Greg Hackmann, Fei Sun Structural Engineering: Nestor Castaneda, Shirley Dyke 24

25 For More Information G. Hackmann, F. Sun, N. Castaneda, C. Lu, and S. Dyke, A Holis>c Approach to Decentralized Structural Damage Localiza>on Using Wireless Sensor Networks, RTSS Sozware Release: hip:// 25