Design of Wireless Sensor Units with Embedded Statistical ... Stanford Research Law.pdfDesign of...

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


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


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


Initial Development

4

Validation:Alamosa Canyon Bridge(with collaboration with Farrar, et.al. of Los Alamos National Lab)

(Prof. Kiremidjian and Dr. Strasser, 1998)


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)


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



• 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






AnalogSensor

0-5V0111


ADXL210

X

Y



Data Memory Buffer





• 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






AnalogSensor

0-5V0111


ADXL210

X

Y



Data Memory Buffer




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


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


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


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)


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


Sensors Mounted to Bridge Girder

Wired PiezotronicsAccelerometer Crowsbow

Accelerometer

Wireless Sensing Units


Data Acquisition

Wireless Data Acquisition Unit

DactronCabled System


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.)


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


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


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


• 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


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)


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


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)


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.


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]


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


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


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


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

Design of Wireless Sensor Units with Embedded Statistical ... Stanford Research Law.pdfDesign of...

Documents

Transcript of Design of Wireless Sensor Units with Embedded Statistical ... Stanford Research Law.pdfDesign of...