Structural Health Monitoring
CSE 520S Fall 2011
Structural Health Monitoring (SHM) Problem: detect and localize damage to a structure Wireless sensor networks (WSNs) monitor at unprecedented temporal
and spaCal granulariCes
Key Challenges: Long-‐term monitoring Rapid on-‐demand analysis Resource and energy constraints
2
Centralized Designs Wisden [Xu, SenSys 2004]
Services for reliable transmission of raw data
Golden Gate Bridge [Kim, IPSN 2007] 46-‐hop network deployed along Golden Gate Bridge
BriMon [Chebrolu, MobiSys 2008] Trains as data mules
Torre Aquila [CerioZ, IPSN 2009] Heterogeneous sensors, most with low data rate
Primarily focus on data transport issues
3
Decentralized Design Principles Raw sensor data is too large to stream back to the
base staCon Damage detecCon is too complex to run enCrely
onboard the sensors
SoluCon: decentralized codesign Select an algorithm which can be run parCally on the
motes Send back (smaller!) parCal results to the base staCon to
complete computaCon
4
Raw Data
Decentralized System Evolution Damage LocalizaCon Assurance Criterion (DLAC)
No collaboraCon needed among nodes => lightweight network architecture
Some limitaCons in damage detecCon
Flexibility-‐Based Methods CollaboraCon among nodes => more complex architecture, but
more robust damage localizaCon Even more energy savings through sensor selecCon
5
Damage Localization Assurance Criterion (DLAC) Collect vibraCon data and use to idenCfy structure’s
natural frequencies [Messina, J. Sound and Vibra:on, 1998] “Signature” of structure’s health
Several traits useful for a decentralized system No data exchanged among nodes IniCal stages are computaConally inexpensive Later stages have much smaller inputs (typically <1% of iniCal data
set)
6
Vibration Data Input Damage LocalizaCon Algorithm
7
0 1 2 3 4 5 6 7 8!2000
!1500
!1000
!500
0
500
1000
1500Time History WS2
Time(s)
Am
plit
ud
e
0 10 20 30 40 50!20
0
20
40
60
80
100
120
140Power Spectrum WS2
Frequency (Hz)
Am
plit
ud
e(d
B)
FFT + Power Spectrum Analysis Damage LocalizaCon Algorithm
8
Curve Fitting Damage LocalizaCon Algorithm
9
! "! #! $! %! &!!#!
!
#!
%!
'!
(!
"!!
"#!
"%!)*+,-./0,12-34.5/#
6-,73,819.:;<=
>40?@23A,:AB=
.
.
)*+,-./0,12-34
C3-D,.6@22@8E
DLAC A mathemaCcal 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 locaCons (Possibly) damaged structure
10
DLAC Output Damage LocalizaCon Algorithm
11
0 5 10 15 200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
DLAC WS2
Element Position
Highest correlaCon to damage at LocaCon 5
D: # of samples P: # of natural freq. (D » P)
Data Flow Analysis Damage LocalizaCon Algorithm
12
(1) FFT
(2) Power Spectrum
(3) Curve FiZng
(4) DLAC
D Integers
Healthy Model Damaged LocaCon
D Floats
D/2 Floats
P Floats
Data Flow Analysis Damage LocalizaCon Algorithm
13
(1) FFT
(2) Power Spectrum
(3) Curve FiZng
(4) DLAC Healthy Model Damaged LocaCon
8192 Bytes
4096 Bytes
D: 2048 P: 5 Integer: 2 bytes Float: 4 bytes
4096 Bytes
20 Bytes
EffecCve compression raCo of 204:1
Implementation Hardware plaoorm: Intel/Crossbow Imote2 + ITS400
sensor board 13 – 416 MHz XScale CPU 32 MB ROM, 32 MB SDRAM CC2420 802.15.4-‐compliant radio 3-‐axis accelerometer on sensor board
Sosware plaoorm: TinyOS 1.1 243 KB ROM, 71 KB RAM
14
Evaluation: Truss 5.6 m steel truss structure at UIUC
Fourteen 0.4 m long bays, siZng on four rigid supports
11 Imote2s atached to frontal pane
15
Wireless SensorTruss Frontal Panel
Fig 12. DLAC results for truss bay # 3
6.0 CONCLUSIONS
In this study a successful demonstration for an in-situ experimental validation of a
correlation-based decentralized damage detection technique using a wireless sensor network has
been performed. Structural damage was detected with sufficiently high correlation percentage in
two experimental structures independently of the damage hypothesis used in the sensitivity
matrix. On-board processing iMote2 capacities were exploited to reduce communication load
and make the application scalable within a wireless sensor network.
7.0 ACKNOWLEDGMENT S
Funding for this research is provided in part by the National Science Foundation; grant NSF
NeTS-NOSS Grant CNS-0627126, by Washington University in St. Louis. Additionally, the
authors would like to thank Prof. Bill Spencer and Shin-Ae Jang for the use of and assistance
with the experimental truss.
8.0 REFERENCES
Clayton, E.H. (2002), “Development of an Experimental Model for the Study of Infrastructure
Preservation”, Proceedings of the National Conference on Undergraduate Research,
Whitewater, Wisconsin.
Clayton, E.H., Koh, B.H., Xing, G., Fok, C.L., Dyke, S.J. and Lu, C. (2005), “Damage
Detection and Correlation-based Localization Using Wireless Mote Sensors”, Proceedings
of ’05 The 13Th
Mediterranean Conference on Control and Automation, Limassol, Cyprus.
Clayton, E.H. (2006), “ Frequency Correlation-based Structural Health Monitoring with Smart
Wireless Sensors”, Master of Science Thesis, Washington University in St. Louis.
1234567891011120
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 3Y = 0.868
DLAC WS #32
Truss Central Bay Position
1234567891011120
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 3Y = 0.864
DLAC WS #45
Truss Central Bay Position
1234567891011120
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 3Y = 0.871
DLAC WS #67
Truss Central Bay Position
1234567891011120
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 3Y = 0.873
DLAC WS #28
Truss Central Bay Position
1234567891011120
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 3Y = 0.825
DLAC WS #35
Truss Central Bay Position
1234567891011120
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 3Y = 0.865
DLAC WS #75
Truss Central Bay Position
Damage correctly localized to third bay
Latency
0 2000 4000 6000 8000 10000 12000 14000
Decentralized
Centralized
Latency (ms)
Sampling ComputaCon CommunicaCon
EvaluaCon
16
Energy Consumption
0 1 2 3 4 5 6
Decentralized
Centralized
Energy consump3on (J)
Sampling ComputaCon CommunicaCon
EvaluaCon
17
DLAC: Findings Onboard processing reduces latency by 66% and energy
consumpCon by 71% EffecCvely localized damage to discrete locaCons on two
structures
Results indicate the power of holisCc energy management
18
G. Hackmann, F. Sun, N. Castaneda, C. Lu, and S. Dyke, “A HolisCc Approach to Decentralized Structural Damage LocalizaCon Using Wireless Sensor Networks”, RTSS, 2008.
Flexibility-‐Based Methods Structures flex slightly when a force is applied Structural weakening => decreased sCffness Flexibility acts as a “signature” of the structure’s health
Two flexibility-‐based methods of interest Beam-‐like structures: Angles-‐Between-‐String-‐and-‐Horizon
flexibility method (ASHFM) [Duan, J. Structural Engineering and Mechanics 09]
Truss-‐like structures: Axial Strain flexibility method (ASFM) [Yan, J. Smart Structures and Systems 09]
19
θ
Network Architecture Sensors form physically-‐
colocated groups Group members collect raw
vibraCon data and process into power spectrum data
Group leaders collect corresponding power spectrum data from children, correlaCng into modal parameters (natural frequencies + mode shapes)
20
Base Sta3on
Group Leader
Group Leader
Group Member
Group Member
Group Member
Group Member
Group Member
Network Architecture Base sta3on collects modal
parameters from group leaders, completes processing into structural flexibility
Output is compared against “baseline” output from healthy structure
Differences in flexibility can be used to detect and localize damage
21
Base Sta3on
Group Leader
Group Leader
Group Member
Group Member
Group Member
Group Member
Group Member
Standard Data Flow
22
Sensing
FFT
Power Spectrum
Cross Spectral Density
Singular Value DecomposiCon
2 x D ints
D floats
Group Leader
Flexibility
Base StaCon
D matrices
Group Member
D floats P natural frequencies +
mode shapes
D: # of samples P: # of natural freq. (D » P)
Enhanced Distributed Data Flow
23
Sensing
FFT
Power Spectrum
2 x D ints
D floats
Peak Picking
D floats
Cross Spectral Density
Singular Value DecomposiCon
Group Leader
Flexibility
Base StaCon
P matrices
P natural frequencies + mode shapes
P floats
Group Member
D: # of samples P: # of natural freq. (D » P)
Multi-‐Resolution Damage Localization Under ASHFM and ASFM, only a handful of sensors are
needed to detect damage As more sensors are added, localizaCon gets more fine-‐
grained Significant energy savings by exploiCng localized nature of
flexibility-‐based approach
24
Evaluation: Simulated Truss SimulaCon of UIUC truss structure
Simulated sensor data generated in MATLAB and injected into live applicaCon using “fake” sensor driver Intact data set: no damages Damaged data set: three members reduced on les side of truss,
four on right side
Result: Level 1 idenCfied damage on both halves of truss; Level 2 localized damage correctly to all seven members
25
Evaluation: Simulated Truss Codesigned architecture reduces
communicaCon latency from esCmated 87 s to 0.21 s
78.9% of energy atributable to synchronizaCon and sensing
Compare to theoreCcal supply of 20,250 J from 3x AAA bateries
26
Group Member
SynchronizaCon 12.1 J
Sensing 23.0 J
ComputaCon 9.28 J
CommunicaCon 0.08 J
Group Leader
SynchronizaCon 16.2 J
Sensing 21.2 J
ComputaCon 8.52 J
CommunicaCon 0.76 J
G. Hackmann, W. Guo, G. Yan, C. Lu, and S. Dyke, “Cyber-‐Physical Codesign of Distributed Structural Health Monitoring With Wireless Sensor Networks”, ICCPS, 2010.
Preliminary Test Full-‐Scale Truss
27
Image source: Zhuoxiong Sun, Purdue University
Test Results: Full-‐Scale Truss
Two levels of damage localizaCon
Level 1: localized damage to bay 9
Level 2: localized damage to element 42
28
2 3 4 5 6 7 8 910 20 3132 420
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2x 10−6
Truss Element Number
AS F
lexi
bilit
y D
amag
e In
dica
tor
Conclusion Codesign approach integrates two SHM methods with
efficient distributed compuCng architectures Mul:-‐level search strategy only acCvates sensors in area of
interest; many sensors remain asleep Shown to localize damage to real beam and truss
structures
Long-‐term goal: a general codesign framework for integrated sensing and control
29
Papers G. Hackmann, F. Sun, N. Castaneda, C. Lu, and S. Dyke,
A HolisCc Approach to Decentralized Structural Damage LocalizaCon Using Wireless Sensor Networks, IEEE Real-‐Time Systems Symposium (RTSS'08), December 2008.
G. Hackmann, W. Guo, G. Yan, C. Lu and S. Dyke, Cyber-‐Physical Codesign of Distributed Structural Health Monitoring with Wireless Sensor Networks, ACM/IEEE InternaConal Conference on Cyber-‐Physical Systems (ICCPS'10), April 2010.
30
Evaluation: Cantilever Beam 2.75 m x 7.6 cm x 0.6 cm steel beam in
Structural Control and Earthquake Engineering Lab
Damage simulated at three locaCons by ataching a steel bar
7 Imote2s atached at equidistant locaCons
32 0.6
6 m
1.3
5 m
1.9
m
2.7
5 m
Wireless Sensor
Damage Location
damage case by applying a hammer strike along the weaker bending axis. Results reported using
the entire network are depicted in Figs. 6, 7 and 8 where corresponding identified natural
frequencies and DLAC measurements are introduced for each damage scenario. DLAC values
determined at sensors along the length of the beam are provided. Values close to unity indicate
damage location. The entire network report successful damage detection results for all damage
scenarios with correlation measurements greater than 90% at the damaged positions. Recall
experimental damage positions D1, D2 and D3 are associated with elements 5, 10 and 14,
respectively. Despite consistency in the results, some of the sensors report correlation
measurements greater than 50% for some of the element positions. As explained previously,
results of correlation-based methods may not be unique. Frequency change vectors associated
with one damage location could be potentially the same as those obtained with several
combinations of damage location when reduced numbers of modes are used. Therefore, the
inclusion of more modes is expected to clarify the results by concentrating the correlation
measurements around one damage location. Note that these results are obtained with a damage
hypothesis of only 67% of the actual damage. Two additional damage hypotheses are
implemented to test the DLAC performance off-line using different damage assumptions and
acceleration records previously obtained for debugging purposes. New sensitivities matrices and
corresponding frequency change vectors were developed with a prescribed analytical damages
equivalent to 200% and 33% of the actual damage. Results showed the same tendencies and
consistency, and were also successful for all damage scenarios with high correlation
measurements.
Fig 6. DLAC results for element position # 5
0 10 200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 5
Y = 0.94
DLAC WS1
Element Position
0 10 200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 5
Y = 0.971
DLAC WS2
Element Position
0 10 200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 5
Y = 0.972
DLAC WS3
Element Position
0 10 200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 5
Y = 0.955
DLAC WS4
Element Position
0 10 200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 5
Y = 0.964
DLAC WS5
Element Position
0 10 200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 5
Y = 0.965
DLAC WS6
Element Position
0 10 200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X = 5
Y = 0.97
DLAC WS7
Element Position
Damage correctly localized in all three trials
Top Related