benefits of structural health monitoring of bridges · what is structural health monitoring •use...

Benefits of Structural Health

Monitoring of Bridges

Brent Phares

Bridge Engineering Center

Iowa State University

bphares@iastate.edu

What is Structural Health

Monitoring • Use quantitative information

– Sensors, measurements, images, etc.

• To make management decisions, to learn

about condition, or to establish behavior.

• But, why use SHM.

Traditional Evaluation

1968 - National Bridge Inspection

Program initiated

Visual inspection

Conventional Calculations

Basic Testing Concept • Record field measurements on structure subjected to

loads to identify actual structural response.

• Field data may be used to “calibrate” an analytical model

that closely represents the behavior observed in the

field.

• May be used for detecting damage or deterioration,

general structural evaluations, and for developing load

ratings.

Objective: Short-Term SHM • Collect data, over a relatively short period

of time (several hours to one day), for the

purpose of obtaining quantitative

information that allows for:

– Improving qualitative assessments.

– Improving future designs.

– Determining “on-the-fly” modifications (e.g.,

during construction).

Application #1 • Provide structural performance

recommendations for evaluation

– Load distribution

– Stress level

– Serviceability (deflection, shear keys)

• Guidelines for evaluation – X-section strain compatibility

– Deflections within AASHTO limits

– W/ regard to “shear keys” transverse distribution satisfied AASHTO limits

• Decision: Delay replacement

• Supplementary laboratory strength tests

Application #2 • Determine if the excessive wind event and

associated bridge movement that was reported

was “accurate”.

• Install a monitoring system for remote alerting of

future events.

• Short-term load test

– Identify dynamic modal characteristics

• Strain and acceleration

• Bridge/wind interaction possible

• Long-term monitoring

– Girder strain and acceleration, wind speed

and direction

• Data loggers

• 24x7 data collection

• Alert system

– Cellular communications

Application #3

• Evaluate effectiveness of

strengthening systems

– Strain profile

– Stiffness increase

– Change in load distribution

FRP Plates Bolted Angles

Post-tensioning

• Added material (FRP, Steel)

– Short and long-term evaluations generally reveals that

strengthening mechanisms are effective

• Short and long-term increase in stiffness

• Minimal load distribution effect

CFRP PLATE

STRAIN GAGE

TB

EXT

T

hwebhCFRP

Application #4 • Primary: Determine if load rating is reasonable

• Secondary: Validate evaluation process for subsequent permit trucks

900,000 lb permit load (~RF 0.5)

• Tested with combinations of one and two loaded tandem axle dump trucks

• Learned about behavior

– Composite action

– End restraint

– Live load distribution • Improved load distribution characteristics used

in hand calculations; changed RF to 0.9

– Rating w/ validated analytical model • RF ~ 1.0 (permit truck)

BRIDGELC

G1 G2 G4G3 G5 G6

-100

-50

0

50

100

150

200

250

0 100 200 300 400 500 600

Truck Position, ft

Mic

rostr

ain

G2

G3

G4

G5

-100

-50

0

50

100

150

200

250

0 100 200 300 400 500 600

Truck Position, ft

Mic

rostr

ain

G2

G3

G4

G5

Analytical: Experimental:

• Prediction validity for permit truck

Application #5

• Short term – eliminate panel overstress during

construction.

• Long term – monitor redistribution of loads in

hangers due to concrete creep.

• Strain (force) correlated with target design

values during construction

September 8-10, 2004

Application #6 Incremental Launch

of the U.S. 20 Iowa River Bridge

Girders Assembled in Launching Pit

and Supported by Rollers

Substructure Monitoring • General pier behavior

• (drilled shaft and driven

pile)

– Column base strain

– Column base translation

and tilt

– Cap beam tilt

At near and far column faces

near face

far face

S

0

Str

ain

Launch Distance (ft)

Far face

Near face 0

+

-

• Largest day launch cumulative column stress measured

was 600 psi

• Residual stress at end of day launch

Monitoring Results - Substructure

S N

Superstructure Monitoring • Girder load distribution

– Bending

• Cross-frame behavior

• Roller contact stresses

– Bottom flange

– Web

– Flange to web welds

Monitoring Results - Superstructure

Roller

-1200

-1000

-800

-600

-400

-200

0

200

823 823.25 823.5 823.75 824Launch distance (ft)

Str

ain

-7634 με CL of Roller

at Pier 6

• Significant longitudinal flange strain measured > Fy

Longitudinal Strain

Monitoring Results - Superstructure

• Significant vertical strain measured

Roller

-2400

-2000

-1600

-1200

-800

-400

0.00 1.50 3.00 4.50 6.00 7.50 9.00Distance above the bottom of the flange plate (in.)

Str

ain

1.125

-2037 με

-1838 με

Vertical Strain

Member

Type

Design

Force

Calculated

Force (WB1)

Calculated

Force (WB5)

Upper

Chord

20 kips C 42.6 kips T 86.2 kips T

Diagonals 38 kips T

or C

56.2 kips T 172.1 kips T

Bottom

Chord

20 kips T

or C

31.1 kips T 39.7 kips C

Monitoring Results - Superstructure • Cross-frame behavior is complex and sensitive

-axial forces, biaxial bending, and torsion

• Measured values sometimes exceeded design values

Objective: Long-Term SHM

• Assess (engineering based) the condition of a bridge continuously without human interaction over long periods of time (days, months, years)

– “Instantly” detect abnormalities.

– Assess load carrying capability.

– Active bridge management system.

The Need

• Bridge owners are looking for autonomous systems that assess structural integrity/safety/etc.

– Enter structural health monitoring

• Research in this area has been going on for 20+ years…with very, very few successes

– Previous approaches relied upon dynamics

– And even fewer attempted field demonstrations

System Attributes Desired by Owners

• Damage detection

– All types of damage at varying locations

– Autonomous data collection, reduction, evaluation, and storage

– Reporting in real-time and in understandable formats

• Load rating (including better capacity estimate) and remaining life estimates

Our Measurement Metric of Choice

• Strain, strain, strain.

• In particular, strain due to the passage of live loads…specifically, five-axle “semi” trucks

– Heavy weights and large numbers of them on the highway system.

Difficulties with Autonomous SHM

• Unknowns, unknowns, unknowns

– Vehicle weight and geometry

– Traffic density and position

– Dynamic impact and suspension system variability

• Conventional analysis techniques are difficult/impossible to reliably perform

Data, Data, Data

Generic System Architecture

Data

Rating/Capacity Model

Damage Detection

Decision for Bridge Rating

Remaining Life

Decision for Bridge

Inspection

Long-Term Management

Truck detection needed

Truck Detection Methodology

Raw Strain Data

Case #1 Case #2

Case #3 Case #4 Case #5

Case #6 Case #7

Eliminate Concurrent Events: Side by Side

Eliminate Concurrent Events: Same lane

Desired:

Fundamental Situations

North Lane South Lane

DL23 DL24 DL21 DL22

(DL13) (DL14) (DL11) (DL12)

① ② ③ ④ ⑤

1

d12 Lb +Lt

Deck bottom

sensor line 2

Deck bottom

sensor line 1

Truck and Lane Detection Basics

Deck strains

Benefits of using strain rate

• Eliminate global strain

effects

• Clearly shows and exploits

the localized strain effects

of truck wheel loads

• Accentuates the behaviors

of interest

Five Peaks (Axles)

Five Peaks (Axles)

Deck Strain Rates

Axle Detection Using Deck Strain Rate

Capabilities

• Can detect: event occurrence, determine passage lane, determine number of axles, estimated (~+/- 0.5’) axle spacing, etc.

• Eliminate instances where multiple trucks are on the bridge at the same time

• Can be applied to bridges of virtually any configuration.

Video - I-80 Bridge: Truck #1

Lane Detection Using Deck Strain Rates

5-axle truck #1 5-axle truck #2

Lane Detection Using Deck Strain Rates

5-axle truck #1 5-axle truck #2

Damage Detection

Pattern Recognition

• Train the system to recognize and develop relationships between sensors that are indicative of typical/normal performance

• Deviations from trained relationships are indicators that something has changed But what pattern?

Algorithm Fundamentals

• Footprint matching

– Each traffic event leaves a footprint with distinct strain record shape and magnitude(s)

– Need a combination of bridge engineering and statistical expertise to create meaningful relationships/patterns

First Generation System

FINALLY! An autonomous temperature compensation algorithm that works.

Capture the pseudo-static behavior

Limitations of First Generation System

• Limits must be established manually

– Time consuming and subjective

• Notable “spread” in the match relationships

• Sometimes multiple match relationships exist

• Results need some further interpretation

Second Generation System

• Develop and implement an autonomous truck parameter detection sub-system – Travel lane, concurrent vehicle presence, number of

axles, axle spacing, vehicle weight, and speed

– Allows decimation of data to reduce spread and potential for multiple relationships

• Systematic approach for – Establishing normal limits

– Direct interpretation that shows damage occurrence and location

0 20 40 60 80 100 120 140-180

-160

-140

-120

-100

-80

-60

-40

-20

0

B-SG-BF-H (Max, )

C-S

G-C

B(1

)-V

(M

in, )

2axle r

2axle l

3axle r

3axle l

4axle r

4axle l

5axle r

5axle l

6axle r

6axle l

0 20 40 60 80 100-100

-90

-80

-70

-60

-50

-40

-30

-20

-10

B-SG-BF-H (Max, )

C-S

G-C

B(1

)-V

(M

in, )

5axle r

0 20 40 60 80 100 120 140-200

-180

-160

-140

-120

-100

-80

-60

-40

-20

0

B-SG-BF-H (max, )

C-S

G-C

B(1

)-V

(m

in, )

Data Selection

, 0, 1,pred ij ij ij jS a a S

For the undamaged structure, the strain of one sensor can be used to predict the strain of another sensor (counter to most structural analyses)

20 40 60 80 100 120 14020

40

60

80

100

120

140

160

180

B-SG-BF-H (Range, )

C-S

G-C

B(1

)-V

(R

an

ge

n, )

5axle r

Model1 (heavy)

Model2 (all)

Model3 (light)

, ( 0 1 )ij i pred ij i ij ij jR S S S a a S

40 40

, ,1 1

( 1 40)i ij std ij stdj i

CRS R R i to

,

ij ij

ij std

ij

RR

3

3i i i

i i i

UCL

LCL

0 50 100 150 200 250 300-3

-2

-1

0

1

2UCL

LCL

CL

1.6753

-1.6753

Truck Groups

CR

S

C-SG-WG(1)-V Group Size:10 Level0

training

Damage Detection and Localization

• Post-training (i.e., monitoring stage)

– Calculate residual matrix using the training stage prediction models

– Calculate CRS

– Plot data on the control chart

– Out of limit points indicate structural damage

0 50 100 150 200 250 300-3

-2

-1

0

1

2UCL

LCL

CL

Truck Groups

CR

S

C-SG-WG(1)-V Group Size:10 Level3

training

monitoring

Second Generation Validation

• US30 bridge over the South Skunk River

– Two-girder, fracture critical bridge

– Multiple fatigue sensitive locations

Specimen

Top Web Plate

Steel Strut

Stringer

Concrete Abutment Pedestal

Anchored Threaded Rods and Epoxy

Bottom Web Plate

Typical Results

• Training data collected

– 3,653 heavy, right-lane, five-axle trucks

Sensor 3 Sensor 4

Girder Bottom Flange Stringer Bottom Flange

Web-Gap Area 1 Web-Gap Area 2

0 10 20 30 40 50 60 70 80 90 100-6

-4

-2

0

2

4UCL=3.48

LCL=-1.79

CL=0.85

Specimen2-Sensor3 Undamaged

Truck Groups

Rsum

Training

Testing

0 10 20 30 40 50 60 70 80 90 100-3

-2

-1

0

1

2

UCL=1.43

LCL=-2.06

CL=-0.31

Specimen2-B-NG-BF-H Undamaged

Truck Groups

Rsum

Training

Testing

0 10 20 30 40 50 60 70 80 90 100-2

-1

0

1

2UCL=1.73

LCL=-1.58

CL=0.07

Specimen2-C-SG-CB(5)-V Undamaged

Truck Groups

Rsum

Training

Testing

0 10 20 30 40 50 60 70 80 90 100-4

-2

0

2

4

6

UCL=4.14

LCL=-3.8

CL=0.17

Specimen2-Sensor4 Undamaged

Truck Groups

Rsum

Training

Testing

0 10 20 30 40 50 60 70 80 90 100-4

-2

0

2

4

UCL=1.17

LCL=-2.21

CL=-0.52

Specimen2-D-NS-BF-H Undamaged

Truck Groups

Rsum

Training

Testing

0 10 20 30 40 50 60 70 80 90 100-3

-2

-1

0

1

2

3UCL=2.71

LCL=-2.09

CL=0.31

Specimen2-E-SG-CB(5)-V Undamaged

Truck Groups

Rsum

Training

Testing

Fatigue Crack Induced

Sensor 3 Sensor 4

Girder Bottom Flange Stringer Bottom Flange

Web-Gap Area 1 Web-Gap Area 2

0 10 20 30 40 50 60-4

-2

0

2

4

6 UCL=5.65

LCL=-2.05

CL=1.8

Specimen2-Sensor3 Damaged

Truck Groups

Rsum

0 10 20 30 40 50 60-15

-10

-5

0

5UCL=3.44

LCL=-3.86

CL=-0.21

Specimen2-Sensor4 Damaged

Truck Groups

Rsum

0 10 20 30 40 50 60-3

-2

-1

0

1

2

UCL=1.43

LCL=-2.06

CL=-0.31

Specimen2-B-NG-BF-H Damage1

Truck Groups

Rsum

0 10 20 30 40 50 60-3

-2

-1

0

1

2

3

UCL=1.17

LCL=-2.21

CL=-0.52

Specimen2-D-NS-BF-H Damage1

Truck Groups

Rsum

0 10 20 30 40 50 60-2

-1

0

1

2

3

UCL=1.73

LCL=-1.58

CL=0.07

Specimen2-C-SG-CB(5)-V Damage1

Truck Groups

Rsum

0 10 20 30 40 50 60-3

-2

-1

0

1

2

3UCL=2.71

LCL=-2.09

CL=0.31

Specimen2-E-SG-CB(5)-V Damage1

Truck Groups

Rsum

Findings

• Damage of multiple types and multiple levels can be autonomously detected

• A correlation exists between damage level and damage indicator data (remaining life…)

• Unfortunately, algorithm has a higher than expected false-positive rate

Third Generation Improvements

• More robust prediction model – orthogonal regression

-6

-4

-2

0

2

4

6

-6 -4 -2 0 2 4 6

Sen

sor

41

(m

icro

stra

in)

Sensor 25 (microstrain)

Third Generation Improvements

• Damage detection model enhancement

𝒀 = 𝒌𝟏 𝜶𝟏 + 𝜶𝟑𝒙 + 𝒌𝟐 𝜶𝟐 + 𝜶𝟒𝒙

where:

𝑘1= 1 for training data

0 for post training data

and:

𝑘2 = 0 for training data

1 for post training data

𝒀 = 𝜸𝟏 + 𝜸𝟑𝒙

Full Model:

Reduced Model:

𝑭 =𝑹𝑺𝑺 𝒓𝒆𝒅𝒖𝒄𝒆𝒅 − 𝑹𝑺𝑺(𝒇𝒖𝒍𝒍) / 𝒅𝒇𝑹𝑺𝑺(𝒓𝒆𝒅𝒖𝒄𝒆𝒅) − 𝒅𝒇𝑹𝑺𝑺(𝒇𝒖𝒍𝒍)

𝑹𝑺𝑺(𝒇𝒖𝒍𝒍)/𝒅𝒇𝑹𝑺𝑺(𝒇𝒖𝒍𝒍) Model

similarity test:

20

21

22

23

24

25

26

27

28

29

30

80 85 90 95 100 105 110 115 120

Sen

sor

19

(m

icro

stra

in)

Sensor 15 (microstrain)

Training Data Post-Training Data

Linear (Full Model) Linear (Reduced Model)

10

20

30

40

50

60

70

80 85 90 95 100 105 110 115 120

Sen

sor

44

(m

icro

stra

in)

Sensor 15 (microstrain)

Training Data Post-Training Data

Linear (Full Model) Linear (Reduced Model)

No damage

Damage!

Third Generation

• False-positive rate appeared to drop

• All damage still detected

• Additional need: autonomous analysis of F-test results (currently being developed)

– Perhaps a return to control chart concepts!

Fourth Generation

• Statistical strain range damage detection based upon quadruple redundant approach

• Four methods – Strain range

• Single truck

• Group size of ten

– Cross-prediction method

– FSHM

• Fully autonomous data collection, reduction, and analysis software.

Strain Range Method Added

Single truck

Group of ten

Load Rating

Automated Load Rating Determination

• Use bridge response from unknown/partially known trucks.

• Use WIM data to statistically estimate the unknown truck features.

• Benefits: does not require bridge closure to perform load tests, load ratings can be estimated frequently (daily), allows for tracking changes with time.

Hardwired strain gages

Wireless truck position indicator

Engineering based data interpretation

Structural modeling

Model analysis and optimization with field collected data Accurate

Assessment

Note: Automation of a proven approach

Analytical Model Calibration

Load Rating using Calibrated Model

Select a Set of Strain Data with Maximum Strain of 90~100

Structural Health Monitoring System

Sample a Truck from WIM Database with Weight of 70~80 kips, Axle Spacing #3: X ± 2 ft, and Transverse Position of Lane Center ± 2 ft

Load Rating

Distribution

600 Runs

Next ‘Big’ Truck

Other On-going Enhancements

• Capacity estimate from service level measurements.

• Remaining life estimation algorithm.

Future Implementation

Benefits of Using SHM

• No subjectivity in the data and/or

information.

• Can make decisions with greater

confidence.

• Data/information can be collected/used as

frequently as desired (must be cautious to

not get overwhelmed by data volume).

Thank You!

Questions?