design of wireless sensor units with embedded statistical ... stanford research law.pdfdesign of...
TRANSCRIPT
Stanford UniversityKincho H. Law
Design of Wireless Sensor Units with Embedded Statistical Time-Series Damage Detection
Algorithms for Structural Health Monitoring
Jerome P. Lynch, Arvind Sundararajan, Kincho H. Law, Anne S. Kiremidjian, Ed Carryer
The John A. Blume Earthquake Engineering CenterDepartment of Civil and Environmental Engineering
Stanford University, Stanford, CA 94305
Structural Health Monitoring Workshop, UCSDMarch 7, 2003
Collaborators: Dr. Hoon Sohn and Dr. Charles Farrar
Los Alamos National Laboratory
Stanford UniversityKincho H. Law
Research Motivation• Need for a rational and economical structural monitoring
– Highway bridges – 583,000 bridges in the United States• Require federal mandated visual inspections
– Buildings - 6 stories or with total floor areas over 5,500 m2
• Recommend 3 accelerometers (2001 California Building Code & 1997 UBC)
• Variety of sensors needed for civil structures – Accelerometers, strain gages, anemometers
• High Installation & maintenance costs – 25% of total system cost – 75% of installation time (cables)– $27,000 per channel
(Tsing Ma Bridge, HK)
Cost Effective Solution: Hardware + Software
Stanford UniversityKincho H. Law
Future Structural Monitoring Systems• Draw on the available technologies:
– Enhance functionality - local computational power – Lower overall costs - wireless communication between sensors
• Wireless Modular Monitoring System (WiMMS)– Wireless sensing units (computational power included)
– Various communication architectures permissible (peer-to-peer)
CentralizedData Acquisition
SensorsCabling
Sensors
Cables
Data Acq. Unix BoxBus
Sensors
Sensors
Current Conventional Practice
Micro-Processor
WirelessModemSensorsA-to-D
BatteriesWirelessModem PC
CentralizedData Storage
SU
SU
SU
SU
Sensor Units
SU
WirelessCommunication
SU
SM
Wireless Embedded System
Stanford UniversityKincho H. Law
Initial Development
4
Validation:Alamosa Canyon Bridge(with collaboration with Farrar, et.al. of Los Alamos National Lab)
(Prof. Kiremidjian and Dr. Strasser, 1998)
Stanford UniversityKincho H. Law
Wireless Sensing Unit
• Prototype designed and fabricated in 2001
• Off the shelf components used
• Two-layer circuit board designed to house all components
• Size is only 4”x4”x1.5”
• Component Cost is about $400
(Jerome Lynch, 2002)
Stanford UniversityKincho H. Law
Wireless Sensing Unit - Design
• Three channel interface– Single channel 16-bit A/D converter – 10 kHz (maximum)– Two 14-bit digital sensor inputs
• Sensor Transparent– Traditional structural sensors (accelerometers, strain gages, etc)– Environmental sensors (thermal, chemical and biological sensors)
WirelessCommunications
Computational CoreSensing Interface
16-bitAnalog-Digital
Converter(Single Channel)
RISC MicrocontrollerAtmel AVR
AnalogSensor
0-5V0111
Dual Axis (X,Y)MEMS Accelerometer
ADXL210
X
Y
Proxim RangeLan2Wireless Modem
Serial Port Serial Port
Data Memory Buffer
Spread SpectrumEncoding
32-bit Motorola Power PCMPC555
Stanford UniversityKincho H. Law
Wireless Sensing Unit - Design
• Dual processor core – optimized for energy efficiency– Atmel AVR 8-bit microcontroller – low power (8 mA)
• Responsible for overall unit operation– Motorola 32-bit PowerPC microcontroller – high power (100 mA)
• Responsible for engineering analyses• Regularly kept off / turned on by 8-bit processor for analyses
WirelessCommunications
Computational CoreSensing Interface
16-bitAnalog-Digital
Converter(Single Channel)
RISC MicrocontrollerAtmel AVR
AnalogSensor
0-5V0111
Dual Axis (X,Y)MEMS Accelerometer
ADXL210
X
Y
Proxim RangeLan2Wireless Modem
Serial Port Serial Port
Data Memory Buffer
Spread SpectrumEncoding
32-bit Motorola Power PCMPC555
Stanford UniversityKincho H. Law
Wireless Sensing Unit - Design
• Proxim RangeLAN2 wireless modem– 2.4 GHz unregulated FCC ISM radio band– Draws a lot of power – 160 mA active (60 mA in sleep mode)– Frequency hopping spread spectrum – highly reliable– Open space range – 1000 feet– Enclosed range – 500 feet
WirelessCommunications
Computational CoreSensing Interface
16-bitAnalog-Digital
Converter(Single Channel)
RISC MicrocontrollerAtmel AVR
AnalogSensor
0-5V0111
Dual Axis (X,Y)MEMS Accelerometer
ADXL210
X
Y
Proxim RangeLan2Wireless Modem
Serial Port Serial Port
Data Memory Buffer
Spread SpectrumEncoding
32-bit Motorola Power PCMPC555
Stanford UniversityKincho H. Law
MEMS Accelerometers• Micro-electro mechanical system (MEMS)
– Accurate and sensitive– Smaller form factors and lower unit costs– Cost advantage - integration of digital circuitry
• In this study, four MEMS Accelerometers considered– Capacitive architectures:
• Analog Devices ADXL210 • Bosch SMB110 • Crossbow CXL01LF1 (Low “g”
accelerometer) – Piezoresistive architectures:
• High Performance Planar Accelerometer (Partridge et al 2000)
• Laboratory Validation (ADXL210,SMB110,HPPA)• Field Validation (CXL01LF1)
Lightly Implanted Piezoresistor
Proof Mass
Chip Surface
Heavily Implanted Conductors
Stanford UniversityKincho H. Law
180161
180162
180163
18017.84
180175
Stiffness(lb/in)
Mass(lb)
Story
46 cm
1.9 cm 28.5 cm
28.5 cm
28.5 cm
28.5 cm
28.5 cm
152.4 cm
30.5 cm
Laboratory Validation Tests• A 5-DOF aluminum shear model structure
Analog Devices ADXL210
Bosch SMB110
Piezoresistiveaccelerometer
Stanford UniversityKincho H. Law
Structural Response Monitoring & FFT• Sweep sinusoid - A = 0.075” with f = 0.25-3 Hz over 60 seconds• FFT calculated (performed on board by the wireless sensing unit)
0 10 20 30 40 50 60-3
-2
-1
0
1
2
3
Time (seconds)
Acc
eler
atio
n (g
)
SMB110A Measured Absolute Acceleration Response
0 10 20 30 40 50 60-3
-2
-1
0
1
2
3
Time (seconds)
Acc
eler
atio
n (g
)
HPPA Measured Absolute Acceleration Response
0 5 10 15
100
102
Frequency (Hz)
Mag
nitu
deHPPA Frequency Response Function (30Hz)
0 5 10 15
100
102
Frequency (Hz)
Mag
nitu
de
ADXL210 Frequency Response Function (30Hz)
0 10 20 30 40 50 60-3
-2
-1
0
1
2
3
Time (seconds)
Acc
eler
atio
n (g
)
ADXL210 Measured Absolute Acceleration Response
0 5 10 15
100
102
Frequency (Hz)M
agni
tude
Bosch SMB110A Frequency Response Function (30Hz)
2.87 Hz
8.59 Hz 13.5 Hz
f1 = 2.96 Hz
f2 = 8.71 Hz
f3 = 13.70 Hz
Stanford UniversityKincho H. Law
Interfacing of Strain Gages
0 200 400 600 800
-2
0
2
x 10-3
MTS Extentiometer
Strain Time History of Steel Tensile Coupon Test
Str
ain
(in/in
)
0 200 400 600 800
-2
0
2
x 10-3
Time (seconds)
Str
ain
(in/in
)
Strain Gage with WiMMS
-4 -2 0 2 4
x 10-3
-20
-10
0
10
20
Strain (in/in)
For
ce (
kip)
-4 -2 0 2 4
x 10-3
-20
-10
0
10
20
Strain (in/in)
For
ce (
kip)
(Micro Measurement)
Stanford UniversityKincho H. Law
Field Validation - Alamosa Canyon Bridge• Constructed in 1937 (7 simply supported spans)• 6 Steel Girders (W30x116 @ 58” CL) with 6” concrete deck
50'
W30x116
Girder 6
Girder 5
Girder 4
Girder 3
Girder 2
Girder 1
S1
S2 S3,S4
S5
S6S7
S8
CL N
1' 4"8'
1'
4"
15"S3S4
NoYesAnti-aliased Output
-2.5 VOffset at 0g
60 µg0.5 mgRMS Resolution
2000 Hz50 HzBandwidth
1 V/g2 V/gSensitivity
0 + 4 g0 + 1 gMeasurement Range
Piezotronics PCB336Crossbow CXL01LF1Sensor Property
(Collaboration with Los Alamos Nat. Lab.)
WiMMS Dactron System
Stanford UniversityKincho H. Law
Sensors Mounted to Bridge Girder
Wired PiezotronicsAccelerometer Crowsbow
Accelerometer
Wireless Sensing Units
Stanford UniversityKincho H. Law
Data Acquisition
Wireless Data Acquisition Unit
DactronCabled System
Stanford UniversityKincho H. Law
Modal Hammer Test• 12 lb Piezotronics PCB86C50 modal hammer• Impulse at span center, sensor location S3• Four modes identified
– Hammer - 6.7, 8.2, 11.4, 12 Hz
0.9 1 1.1 1.2 1.3 1.4 1.5 1.6-0.18
-0.09
0
0.09
0.18
Time (sec)
Acc
eler
atio
n (g
)
Response of Alamosa Canyon Bridge to Modal Hammer (Los Alamos Lab System)
Piezotronics PCB336A @ 320 Hz
0.9 1 1.1 1.2 1.3 1.4 1.5 1.6-0.18
-0.09
0
0.09
0.18
Acc
eler
atio
n (g
)
Response of Alamosa Canyon Bridge to Modal Hammer (WiMMS System)
Crossbow CXL01LF1 @ 976.56 Hz
Time (sec)
0 5 10 15 20 25 3010
-3
10-2
10-1
100
101
Mag
nitu
de
FRF of Alamosa Canyon Bridge to Modal Hammer
Frequency (Hz)
LANL PCB336 WiMMS CXL01LF1
(Collaboration with Los Alamos Nat. Lab.)
Stanford UniversityKincho H. Law
Dynamic Vehicle Test• 40 mph truck drives over a 2x4 wood stud• Acceleration at sensor location S7• Four modes identified
– Van - 6.8, 8.1, 11.2, 11.9 Hz– Hammer - 6.7, 8.2, 11.4, 12.0 Hz
0 1 2 3 4 5 6 7 8 9 10
-0.1
0
0.1
Time (sec)
Acc
eler
atio
n (g
)
Response of Alamosa Canyon Bridge to Van Excitation (Los Alamos Lab System)
Piezotronics PCB336A @ 320 Hz
0 1 2 3 4 5 6 7 8 9 10
-0.1
0
0.1
Time (sec)
Acc
eler
atio
n (g
)
Response of Alamosa Canyon Bridge to Van Excitation (WiMMS System)
Crossbow CXL01LF1 @ 244.14 Hz
0 5 10 15 20 25 3010
-3
10-2
10-1
100
101
Mag
nitu
de
FRF of Alamosa Canyon Bridge to Modal Hammer
Frequency (Hz)
LANL PCB336 WiMMS CXL01LF1
Stanford UniversityKincho H. Law
Ambient Vibrations from I25
0 5 10 1510
-3
10-2
10-1
Mag
nitu
de
Averaged FRF of Alamosa Canyon Bridge to Ambient Vibrations (WiMMS System)
Crossbow CXL01LF1 @ 30 Hz
Frequency (Hz)
Trucks driving south along I25
• Ambient excitations from the I25 bridge adjacent to the ACB
• First three modes identified –6.5, 8.8 and 11.9 Hz
80 100 120 140 160 180 200-8
-4
0
4
8
Time (sec)
Acc
eler
atio
n (g
x 1
0-3) Response of Alamosa Canyon Bridge to Ambient Excitations (WiMMS System)
Crossbow CXL01LF1 @ 30 Hz
1 2 3TRUCK TRUCK TRUCKCAR CAR
I25
ALAMOSA CANYON BRIDGE
Stanford UniversityKincho H. Law
Power Sources• Batteries are used as a primary power source for the wireless sensing
units
• Duty-cycle usage can substantially extend the battery life for the wireless sensing units– Units are normally kept in sleep mode– Returned to an active mode based on a regular schedule– Awakened by triggering events (for example - threshold
accelerations)• Auxiliary power source possible
5 hours15 hours160RangeLAN2 Active
25 hours40 hours60RangeLAN2 Asleep
5 hours15 hours160AT90S8515 Circuit and MPC555 Active
30 hours50 hours54AT90S8515 Circuit and MPC555 Asleep
5-AA E91 (Zn/MnO2) Battery Pack (7.5 V)
5-AA L91 (Li/FeS2) Battery Pack (7.5 V)
Current (mA)
Operational State
Stanford UniversityKincho H. Law
• Motorola PowerPC Microcontroller Core
• 32-bit architecture clocked at 40 Mhz
• Floating point calculations in hardware
• Sufficient Memory -- 448 Kbytes ROM, 26 Kbytes RAM
7.5 V Lithium battery packs
RangeLAN2 wireless modem
PowerPC core
Wireless Sensing Unit with PowerPC
WiMMS
Stanford UniversityKincho H. Law
Embedded Computational Power
• Energy efficient wireless sensor networks– Avoid wireless transmission of
raw time histories (in real time)– Local execution of embedded
engineering analyses• Current embedded algorithms
– Statistical AR-ARX time series damage detection
– Fast-Fourier transform (FFT)Number of Data Points (N)
Per
cent
age
of W
irele
s E
nerg
y 100 %
Interna
l Mem
oryExter
nal M
emory
1600 Points25% (75% Savings)
4096 Points70% (30% Savings)
Stanford UniversityKincho H. Law
Surface Effect Fast Patrol Boat
(Courtesy of Farrar and Sohn)Damage Detection
A,B
F
J
E, F
EG
G
D
B
A
C
C,DK
H
I
H, I
K J
Stanford UniversityKincho H. Law
Auto-Regressive Models Fitted
MATLABSensingUnit
Coef.
0.08910.0891b10
0.14980.1498b9
0.59890.5989b8
1.26461.2646b7
1.52121.5212b6
0.67560.6756b5
0.11710.1171b4
0.47520.4752b3
1.30091.3009b2
1.66501.6650b1
MATLABSensingUnit
Coef.
b20
b19
b18
b17
b16
b15
b14
b13
b12
b11
b10
b9
b8
b7
b6
b5
b4
b3
b2
b1
0.01340.0134
0.23600.2360
0.23970.2397
0.19120.1912
0.22270.2227
-0.0715-0.0715
-0.1971-0.1971
-0.3261-0.3261
-0.5422-0.5422
-0.4464-0.4464
-0.2473-0.2473
-0.1833-0.1833
0.14940.1494
0.53920.5392
0.83770.8377
0.27540.2754
0.06240.0624
0.52930.5293
1.24151.2415
1.59151.5915
AR(10) AR(20)
Stanford UniversityKincho H. Law
Data Compression• Lossless : Data compression without loss in data
integrity.• Lossy : Data compression with “reasonable” errors in
data reconstruction.
Lossless Data CompressionOutput Code StreamData Points Buffer
Entropycoder
Example : Static Huffman coder
0.000.000.0083.00Lossless
ARXMSE(5,5)
ARMSE(30)
DataMSE
Compression Ratio(%)
Scheme
16 Bit High Speed Boat data histories used in the simulation. Results averaged over three different time histories.
Stanford UniversityKincho H. Law
Lossy Compression
Data PointsBuffer
Wavelet Transform Filter Bank
EntropycoderQuantizer
• A Daubechies-4 Discrete wavelet transform used• A Uniform Quantizer : xq = round(x/q)
*H 2
*G 2
*G 2
*H 2Input x[n]Stream
D1[n]
D2[n]
A2[n]
Stanford UniversityKincho H. Law
Preliminary Results
0.000.000.0083.00Lossless
10-710-610-865.60Lossy 1*
10-610-410-758.75Lossy 2*
ARXMSE(5,5)
ARMSE(30)
DataMSE/mean
Compression Ratio(%)Scheme
*Lossy 1 and Lossy 2 represent the results at two different levels of Quantization
Stanford UniversityKincho H. Law
Damage Detection and Assessment Methodologies• System Level Screening
– Rapid assessment – determine presence of damage– Statistical pattern recognition techniques– Embeddable Algorithms
• Damage Diagnosis (Detection)– Damage identification - locate damage regions– Experimental modal analysis or strain-based measurements– Statistical model-based system analysis (Modal vs. Ritz vectors, Ref:
Sohn, H., PhD Thesis)– Data synchronization (Ref: Lei, Kiremidjian, et.al.)
• Damage Inspection– Visual or localized experimental methods
• Damage Prognosis and Self Diagnostics– Estimate the performance level and remaining life– Integrating sensing, self-excitation and control strategies
Stanford UniversityKincho H. Law
Concluding Remarks• An Overview of Our Research in Structural Health Monitoring
– Development of a Wireless Modular Monitoring System– Investigation of Statistical-Based Damage Detection Techniques
• Structural Health Monitoring is a Multidisciplinary Research Problem– Structural mechanics, signal processing, sensing systems,
statistical pattern recognition, computer hardware, telecommunications, embedded computing, etc….
• Future Research– Multiple sensors per module– Integrating monitoring with damage assessment– Integrated monitoring network– Decentralized sensing, monitoring, control– Environmental and other effects
PLENTY OF WORKS TO BE DONE
Stanford UniversityKincho H. Law
Acknowledgements• This research has been partially supported by the Civil and
Mechanical Systems Program, National Science Foundation, Grant Numbers CMS-95261-2, CMS-9988909, and CMS 0121842.
• Collaboration with Dr. Charles Farrar and Dr. Hoon Sohn of Los Alamos National Laboratory is appreciated. The experiment at the Alamosa Canyon Bridge was supported by the Los Alamos Laboratory Directed Research and Development (LDRD) fund.
• Collaborative work with Prof. Anne Kiremidjian, Prof. Tom Kenny and Dr. Ed Carryer. Special credits are attributed to Dr. Jerome Lynch, Dr. Hoon Sohn and Mr. Arvind Sundararajan.
• Any opinions, findings, and conclusions or recommendations expressed in this material are those of the presenter and do notnecessarily reflect the views of the sponsors.
Acknowledgement
Stanford UniversityKincho H. Law