Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

download Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

of 10

Transcript of Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

  • 8/13/2019 Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

    1/10

    VOL. 4, NO. 3, MAY 2009 ISSN 1819-6608

    ARPN Journal of Engineering and Applied Sciences

    2006-2009 Asian Research Publishing Network (ARPN). All rights reserved.

    www.arpnjournals.com

    VIBRATION-BASED STRUCTURAL HEALTH MONITORING

    TECHNIQUE USING STATISTICAL FEATURES FROM

    STRAIN MEASUREMENTS

    A. P. Adewuyi and Z. S. WuDepartment of Urban and Civil Engineering, Ibaraki University, Hitachi, Japan

    E-Mail: [email protected]

    ABSTRACTA statistical vibration-based damage identification algorithm to assess the stability of the measurement data,

    detect and locate damage in civil structures, where variability in response and modal parameters due to measurement noiseand environmental influence is often inevitable, is presented in this paper. The method exploits the regression analysis ofpeak values of the magnitudes of frequency response function (FRF) of target sensors relative to the reference wherein thestatistical features are employed for data reliability assessment and damage localization. Through experimentalinvestigation of a flexural structure using conventional strain gauges and long gauge fiber Bragg gratings (FBG) sensors,the importance of the technique for civil SHM is established and presented in an easy-to-interpret graphical format foreffective implementation of results. The statistical approach is very effective for damage localization using strain data.

    Keywords: damage identification, statistical features, measurement noise, dynamic strain, FBG sensors, structural health monitoring.

    INTRODUCTIONProgressive deterioration of civil infrastructure

    begins once they are built and subjected to normalcontinuous and occasional excessive loading, and adverseenvironmental conditions. Because the intrinsic changesthat adversely affect the immediate or future performanceof a structural system are usually unknown a priori,prompt and intensive monitoring of structural systems isvery important. The extension of vibration-based damageidentification (VBDI) techniques to civil infrastructure isincreasingly receiving attentions because of rise in socialand economic demands for structural integrity and safety.During the past few decades, a significant amount ofresearch has been conducted in the area of nondestructivedamage detection via changes in modal responses of astructure. A number of model-free damage identificationtechniques have been developed and successfullyinvestigated primarily because of they are computationallysimple without any need for an updated numerical model.Most notable among these techniques are based on naturalfrequency, mode shape curvatures, modal flexibility andits derivatives, modal stiffness, modal strain energy,frequency response function and its curvature, and powerspectral density (PSD). Physical quantities most relevantand sensitive to the structural properties of interest areselected for monitoring purposes. Extensive literaturereviews and advances on VBDI have been reported byDoebling et al. (1996), Sohn et al. (2003) and Carden andFanning (2004). Li and Wu (2008) and Wu and Li (2007)also develop modal macro-strain vector (MMSV) methodof damage identification using the modal parametersextracted from dynamic macro-strain data recorded byarrays of long-gauge FBG sensors distributed throughoutthe full or some partial areas of the flexural structures.

    Despite the numerous achievements made so far,the search for more reliable strategies for both damagelocalization and quantification in large-scale civil

    structures is still in progress due to inevitable variability inmeasurement data as a result of measurement error andenvironmental factors. So, the ability to correctlydistinguish changes in the modal properties caused bydamage from those arising from variations in themeasurements induced by electrical noise and otherenvironmental conditions is an essential issue requestingserious attention for successful civil SHM. Thus, theprocess of vibration-based SHM can be fundamentallyconsidered as that of statistical pattern recognition. Inquest for statistical damage identification algorithms thatwould take into consideration all the aforementionedfactors, Pape (1993) proposes a technique to identifydamaged parts using statistical methods and measurednatural frequencies. Damages are detected by assessingresonant frequencies that fall outside the mean standarddeviations. The shortcoming of the approach is its inabilityto detect smaller defects. Brincker et al. (1995) use astatistical analysis method independent of the knowledgeof the input signal to detect damage in two concrete beamswith different reinforcement ratios using changes in themeasured vibration frequencies. The method is based onsignificance indicator for any modal frequency or dampingratio determined by scaling the observed changes by theestimated standard deviation of the parameters or a unifiedsignificance indicator expressed as the summation offrequency and damping significance indicators overseveral measured modes. It was noted that measuredmodal parameters with high confidence are weighted moreheavily in the indicator function. Similar to otherfrequency-based damage identification methods, thesignificance indicator proved to be a sensitive structuraldamage indicator, but incapable to localize damage.Doebling and Farrar (1997) develop a statistical procedurethat uses measured coherence functions to estimate theuncertainty bounds on the magnitude and phase of the

    38

  • 8/13/2019 Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

    2/10

    VOL. 4, NO. 3, MAY 2009 ISSN 1819-6608

    ARPN Journal of Engineering and Applied Sciences

    2006-2009 Asian Research Publishing Network (ARPN). All rights reserved.

    www.arpnjournals.com

    frequency response functions (FRFs), and prorogate theuncertainty to modal parameters.

    In this study, a statistical VBDI algorithm based

    on the regression analysis of peak values of themagnitudes of FRF of target sensors relative to thereference is presented for data reliability assessment anddamage localization. The proposed method is suitable forcivil SHM, where variability in response and modalparameters due to measurement noise and environmentalinfluence is often inevitable, to assess data quality prior todamage identification. Also, the ability to effectivelymanage and make sense of enormous amounts of datacollected under continuous monitoring process for aneffective diagnostic and/or prognostic system is an addedadvantage. The effectiveness of the method is establishedthough experimental investigations on a flexural structure

    by using strain gauges and long-gauge fiber Bragg grating(FBG) strain sensors.

    VIBRATION-BASED STRUTURAL HEALTH

    MONITORING USING STATISTICAL FEATURESFarrar et al. (1999) describe the process of SHM

    as fundamentally one of statistical pattern recognition. Asuccessful civil SHM program involves selection andplacement of sensors suitable for measurement of keyparameters that influence the performance and health ofthe structural system. This section presents a vibration-based damage identification algorithm using statisticalfeatures extracted from continuous monitoring of dynamic

    measurement data for a reliable practical application tocivil infrastructure.

    Theoretical backgroundFrom experimental modal analysis (EMA), the strain

    frequency response functions (FRFs), between

    the applied force and measured strain responses can beexpressed as

    )(djkH

    +

    =r rrrr

    krjr

    jkiM

    H)2(

    .)(

    22

    (1)

    where jr represents the

    th

    r strain mode shaperespectively at response measurement point ; andj kr is

    the mode shape component of displacement mode rat

    excitation point at DOF. is the modal mass andthk rM

    r is the modal damping ratio.

    Additionally, the magnitude of strain FRF at resonance forany mode is given by

    jr

    rrr

    kr

    r

    d

    jk MH

    == 22)( (2)

    It is obvious from equation (2) that for a given mode, the

    quantity22 rrr

    kr

    M

    is a constant for all response

    measurements at different locations on the structure due to

    the same applied excitation at DOF.thk

    Most damage identification techniques aredeveloped on the assumption of linear structural behaviorin the pre- and post-damage states. Therefore, dynamicresponses are very suitable for modal parametersuperposition applications. The frequency responsefunctions can be obtained through fast Fourier transforms(FFT). By considering measurement points for thedisplacement transducers and strain sensors at any givenmode, the peak values of FRF magnitude of thedisplacement response data referred to as modal strainvector (MSV) extracted from the peak values of FRFmagnitude of the strain response data is given by

    m

    { }Tmrjrrr ......MSV 21= (3)

    The strain mode shape at a particularmeasurement point relative to a reference sensor of same

    type mounted at an undamaged location is linear and ofconstant slope. Any deviation in this feature is anindication of changes in the dynamic properties of thestructure due to damage. The reference strain sensors

    denoted by refr should be carefully selected from any

    of the sensors located at points with the leastpossibility of damage. In flexural structures, for example,the reference sensor could be placed at regions of lowbending moment.

    m

    Because measurement data can be measuredunder different conditions, the ability to normalize the databecomes very central to the damage detection process. Inthis case, the MSV is normalized by the modal response ofthe reference sensor. This approach is meant to minimizeto the extent possible the variability in the data acquisitionprocess. Thus, the normalized MSV with respect to thereference sensors is given by:

    { }

    =

    refr

    mr

    refr

    mr

    refr

    r

    refr

    rmrrmrr

    )1(21)1(21 ...... (4)

    For long-gauge FBG sensors, the mathematical

    expressions for conventional strain measurement are stillvery applicable except that the modal strain now becomesmodal macro-strain (MMS). Further information onpackaging and performance of long-gauge FBG sensor can

    be obtained from Li and Wu (2007). The developed

    sensors are characterized by the capability of obtaining themeasurements by integrating both local and globalinformation due to the fact that strain is a typical localresponse and distributed sensor placement helps to record

    39

  • 8/13/2019 Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

    3/10

    VOL. 4, NO. 3, MAY 2009 ISSN 1819-6608

    ARPN Journal of Engineering and Applied Sciences

    2006-2009 Asian Research Publishing Network (ARPN). All rights reserved.

    www.arpnjournals.com

    the data covering the large region of a structure. Moreover,distributed strain sensing technique also reduces thenumber of unknown parameters for effective analysis of

    measured data.

    Data interpretation and features extractionThe study of data features required to distinguish

    the damaged structures from undamaged ones has receivedconsiderable attention in the technical literature. AnySHM monitoring system will produce an enormousamount of data from which it will be necessary to selectthe appropriate information. An innovative analysis ofmeasured data and accurate interpretation of extractedfeatures possess essential ability that makes sense of theenormous amounts of data collected during monitoring

    exercise for an effective diagnostic and/or prognosticsystem. This supplies useful information that ismeaningful to bridge owners, inspectors and engineers for

    practical implementation and effective management ofcivil infrastructure. A 3-step damage identificationalgorithm that evaluates the stability of measurement dataand localizes damage in flexural structures using statisticalfeatures is summarized in Figure-1. Inherent in the featureselection process are the ability to identify theperformance of sensors and effectively condense data. Inaddition, provisions are made for automatic updating ofprevious data if the features remain unchanged, whiledamages are localized at the sensor locations characterizedby feature changes.

    Figure-1. Flowchart for structural health monitoring program using statistical features.

    The 3-step algorithm for data interpretation andfeature extraction based on dynamic measurement data

    from different types of sensing techniques as presented inFigure-1 is described as follows:

    Statistical features (slopes of lines

    Data condensation and

    storage

    Step 2Selection and placement of areference andNS target sensors

    Featureschange?

    Y

    N

    Implementation andManagement

    Step 3

    Are datareliable?

    Data acquisition

    N

    Y

    Step 1

    Select peak values of FRF

    magnitude,|H| eak

    Extract modalparameters

    H eakof SRef

    |H|p

    eakofS3

    H eakof SRef

    |H|p

    eakofSNS

    H eakof SRef

    |H|p

    eakofS1

    H eakof SRef

    |H|p

    eakofS2

    GUI: measurements over a eriod of time

    Continuous monitoring

    |H| eakof SRef

    |H|p

    eakofS3

    |H| eakof SRef

    |H|p

    eakofSNS

    H eakof SRef

    |H|p

    eakofS1

    H eakof SRef

    |H|p

    eakofS2

    GUI: continuous measurements at two different times

    Correlated

    ?

    N

    Y

    Assess the stability of measurementthrough statistical correlation of data

    Results interpretationand recommendations

    Step 2

    40

  • 8/13/2019 Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

    4/10

    VOL. 4, NO. 3, MAY 2009 ISSN 1819-6608

    ARPN Journal of Engineering and Applied Sciences

    2006-2009 Asian Research Publishing Network (ARPN). All rights reserved.

    www.arpnjournals.com

    (1) Selection and placement of adequate

    reference sensor and target sensors ( =1, 2

    ) suitable for measuring key parameters that

    influence the performance and health of the structuralsystem are made. Modal parameters are extracted for eachmeasurement after satisfactory preliminary assessment ofthe reliability of the data acquisition systems. The peakvalues of the magnitude of the FRF for each measurementdata are obtained in a way similar to the construction ofeigenvector in modal analysis.

    RefS NsS Ns

    m

    (2) From the records of several measurement dataat a particular state of the structure, a graphical userinterface is created, which consists of m chartscorresponding to the plots ofpeak values of the magnitudeof the FRF of target sensors at y axis against that of thereference sensor at x axis. The consistency of themeasurement data obtained from different sensingtechniques can be assessed through the statisticalcorrelations for each structural condition. For continuousmeasurements at any two different times, comparativeplots ofpeak values of FRF magnitudesare made to assessthe variations in the slopes of lines of fit. Strain as a verysensitive local response can correctly localize damages atsensor points by comparing two consecutive slopes oflines of fit.

    (3) If the slope increases, damage is localized atthe region of the sensor. No change in structural conditionis indicated by constant slope, while slope reduction is animplication of damage at the location of the referencesensor. The deviation of slope of the fit lines from the

    initial value is clearly a statistical feature for damageidentification. Condensation of huge data is guaranteed bythe technique and the data storage devices are more easily

    and effectively managed. Subsequent data collected fromundamaged regions conveniently replace the previous, andthe time-history data and important features immediatelybefore and after damage are also saved. The results areeasy to interpret and appropriate recommendations can be

    made for effective management and implementation.

    EXPERIMENTAL INVESTIGATION

    Experimental set-up and data collectionTo demonstrate the proposed technique

    experimentally, three 503960 mm3 simply supportedsteel beam specimens shown in Figure-2 were used. To

    contrast FE model, the beam was modeled by 16 elements,17 nodes and 32 degrees of freedom. The Youngsmodulus and the density of the specimens are 2.101011

    N/m2and 7862 kg/m3respectively. Damage was simulatedby introducing width reductions from 50mm to 24mm forsingle damage at element 6 and double damages atelements 6 and 10. The intact, single and double damagecases are respectively denoted as C1, C2 and C3.

    Figure-2. Experimental specimens and sensor placement.

    A succession of single-point impact loads wasapplied at arbitrary locations on the beams to simulateloading with a high bandwidth. Such tests were performed15 times and the measured data were used for modalidentification. The dynamic responses were collectedusing four long-gage FBG sensors (F1~F4) of 240mmgauge length attached to the bottom surface of thespecimens and four conventional strain gauges (S1~S4) of5mm gauge length for comparison. In order to compare

    strain measurements between conventional strain gaugesand the long-gauge FBG sensors, the two types ofinstruments were systematically installed such that theircenters coincide.

    The dynamic responses from FBG sensors andstrain gauges were recorded at a sampling rate of 250Hzand 500Hz, respectively; and impulsive excitations byimpact hammer recorded at 500Hz. The FRF was obtainedfrom the spectral densities of the force input and response

    50

    50

    50

    C3: Beam with two damages at elements 6 and 10

    3

    C1: Intact beam

    C2: Beam with one damage at element 6

    41

  • 8/13/2019 Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

    5/10

    VOL. 4, NO. 3, MAY 2009 ISSN 1819-6608

    ARPN Journal of Engineering and Applied Sciences

    2006-2009 Asian Research Publishing Network (ARPN). All rights reserved.

    www.arpnjournals.com

    outputs at each measurement point. The FRF can becalculated via the classical fast Fourier transforms (FFT)algorithms expressed as

    )(

    )(

    )()(

    )()()(

    ff

    fz

    S

    S

    FF

    FZH ==

    (5)

    Where )(Z and )(F are the FFTs of the response

    and force time domain signals, and , and the

    asterisk denotes the complex conjugate.

    )(tz )(tf

    )(zfS is the

    cross spectrum between the output and input signals; and

    )(ffS is the auto-spectrum of the input signals. The

    symbols and denote the response signals and

    excitation signals, respectively. The approach wasemployed to reduce the effect of noise due to uncorrelatedexcitation.

    z f

    Mode shapes were extracted using global curve-fitting techniques. Figure-3 shows that the mode shapescomputed using ANSYS program and measured usingFBG sensors have good agreements.

    0

    0.2

    0.4

    0.6

    0.8

    1

    1.2

    F1 F2 F3 F4

    Sensor designation

    (a)

    Strain

    mod

    e

    shape

    Theory

    Test

    C1

    0

    0.2

    0.4

    0.6

    0.8

    1

    1.2

    F1 F2 F3 F4

    Sensor designation

    (b)

    Strainmo

    deshape Theory

    Test

    C2

    0

    0.2

    0.4

    0.6

    0.8

    1

    1.2

    F1 F2 F3 F4

    Sensor designation

    (c)

    Strainmo

    deshape

    Theory

    Test C3

    Figure-3. Theoretical versus experimental distributed strain mode shape for first mode.

    DISCUSSION OF EXPERIMENTAL RESULTSTable-1 gives the first modal frequency extracted

    from the response data of the three sensors for thedifferent structural states. All the data perfectly agree with

    high precision.

    Table-1. Identified resonant frequencies using different

    sensors.

    Resonant frequencies (Hz)Scenarios

    FBG sensors Strain gauges

    C1 6.021 6.021

    C2 5.822 5.825

    C3 5.638 5.637

    The frequency spectra for the different measurement dataunder a typical excitation for the three structural cases C1,C2 and C3 measurement data are shown in Figures 4 and5.

    42

  • 8/13/2019 Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

    6/10

    VOL. 4, NO. 3, MAY 2009 ISSN 1819-6608

    ARPN Journal of Engineering and Applied Sciences

    2006-2009 Asian Research Publishing Network (ARPN). All rights reserved.

    www.arpnjournals.com

    Figure-4. Typical frequency response spectra from strain gauges.

    0 25 50 75 100 1250

    0.5

    1

    1.5

    2

    2.5

    3

    3.5

    FRFMagnitude(/N)

    Frequency (Hz)

    FBG Sensor - C1

    F1

    F3F2

    F4

    (a)

    5 5.5 6 6.5 70

    0.5

    1

    1.5

    2

    2.5

    3

    3.5

    FRFMagnitude(/N)

    Frequency (Hz)

    F1

    F3F2

    F4(b)

    C1

    5 5.5 6 6.5 70

    1

    2

    3

    4

    5

    6

    FRFMagnitude(/N)

    Frequency (Hz)

    F1

    F3

    F2F4

    C2

    (c)

    5 5.5 6 6.5 70

    2

    4

    6

    8

    10

    FRFMagnitude(/N)

    Frequency (Hz)

    F1

    F3

    F2F4

    C3

    (d)

    0 25 50 75 100 1250

    1

    2

    3

    4

    5

    FRFMagnitude(/N)

    Frequency (Hz)

    Strain Gauge - C1

    S1

    S2

    S3

    S4

    (a)

    5 5.5 6 6.5 70

    0.5

    1

    1.5

    2

    2.5

    FRFMagnitude(/N)

    Frequency (Hz)

    S1

    S2

    S3

    S4

    (b)

    C1

    5 5.5 6 6.5 70

    2

    4

    6

    8

    FRFMagnitude(

    /N)

    Frequency (Hz)

    S1

    S2

    S3

    S4

    C2

    (c)

    5 5.5 6 6.5 70

    2

    4

    6

    8

    10

    12

    FRFMagnitude(/N)

    Frequency (Hz)

    S1

    S2

    S3

    S4

    C3

    (d)

    Figure-5. Typical frequency response spectra measured with long-gage FBG sensors.

    By taking the first modal parameters intoconsideration, it is evident from the figures that the naturalfrequencies and the damping ratios reduce with increase inthe damage extent. The stability of the measurements can

    be evaluated through the statistical correlation of the peakvalues of FRF magnitudes of the target sensors relative tothat of the reference sensor. The reference sensors for thisinvestigation are strain gauge S4 and FBG sensor F4.

    43

  • 8/13/2019 Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

    7/10

    VOL. 4, NO. 3, MAY 2009 ISSN 1819-6608

    ARPN Journal of Engineering and Applied Sciences

    2006-2009 Asian Research Publishing Network (ARPN). All rights reserved.

    www.arpnjournals.com

    Figures 6 and 7 show the behavior of the modalparameters extracted from measurement data collectedfrom the two sensor types under the random excitations of

    different magnitudes for the three structural conditions ofthe beam specimens. From the plots of the modalparameters of each sensor against the reference sensor, the

    linearity of the relationships for all excitation cases withsingle point impulsive load arbitrarily applied on thebeams can be established for all the sensing techniques.

    By fitting the discrete points, good lines of fit implyingstability of the measurements can be obtained.

    Figure-6. Modal strain parameter extracted from strain gauges.

    0

    20

    40

    60

    80

    0 4 8 12 16 20

    Microstrain of reference sensor S4

    M

    M

    So

    fStrain

    Gauges

    C3S1

    C3S2

    C3S3

    C3S4

    0

    3

    6

    9

    12

    0 1 2 3 4 5

    Microstrain of reference sensor S4

    M

    M

    So

    fStrain

    G

    auges

    C1S1

    C1S2

    C1S3

    C1S4

    0

    10

    20

    30

    40

    50

    0 2 4 6 8 10 12

    Microstrain of reference senso r S4

    MMS

    ofStrainGauges

    C2S1

    C2S2

    C2S3

    C2S4

    0

    2

    4

    6

    8

    0 1 2 3 4

    MMS of reference s ensor F4

    MMS

    ofFBGs

    ensors

    C1F1C1F2C1F3C1F4

    0

    5

    10

    15

    20

    25

    0 2 4 6 8

    MMS of reference sensor F4

    MMSofFBGs

    ensors

    C2F1

    C2F2

    C2F3

    C2F4

    0

    15

    30

    45

    60

    0 5 10 15 20

    MMS of reference sensor F4

    MMS

    ofFBGs

    ensors

    C3F1C3F2C3F3C3F4

    Figure-7. Modal macro-strain measurements from FBG sensors.

    44

  • 8/13/2019 Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

    8/10

    VOL. 4, NO. 3, MAY 2009 ISSN 1819-6608

    ARPN Journal of Engineering and Applied Sciences

    2006-2009 Asian Research Publishing Network (ARPN). All rights reserved.

    www.arpnjournals.com

    The stability of the measurement data is evaluated via thecorrelation of the modal data of each sensor type relative

    to its reference as shown in Table-2.

    Table-2.Statistical correlation of modal parameters using different sensors.Correlation coefficients

    CasesStrain gauges FBG sensors

    S1 S2 S3 F1 F2 F3

    C1 0.9992 0.9996 0.9996 0.9964 0.9996 0.999

    C2 0.9993 0.9993 0.999 0.9987 0.9964 0.9979

    C3 0.9994 0.9998 0.9998 0.9997 0.9995 0.9995

    It is clear that the measurement data from all thesensors have perfect statistical correlations with respect to

    the reference. These excellent correlation coefficientssignify the stability and reliability of the response datameasured by all the three sensors. It can be inferred thatthe effects of noise on the measurements is minimal.

    Statistical feature for damage localizationThe lines of fit have dual features: the correlation

    coefficient and the slope or equation of fit line. While theformer assesses the consistency or stability of the data, the

    slope is employed to localize damage. A close monitoringof the variation of slopes of fit lines for each sensorcorresponding to the structural scenarios under different

    excitations at any different times gives clear indication oflocation of damage. The plots of variation of slopes of fitlines for the two sensor types are shown in Figures 8 and9. Table-3 gives the variation of slopes of lines of fit forstrain gauges and FBG sensors in which the underlinedfigures in bold indicate the significant change indicatingdamages.

    Table-3. Statistical features indicated by variation of slopes of fit lines.

    Cases Slope of fit lines

    Strain gauges FBG sensorsS1 S2 S3 F1 F2 F3

    C1 1.145 6.3502 6.0371 0.916 4.738 4.911

    C2 0.9781 14.178 5.8734 1.0209 8.827 5.1164

    C3 0.9917 13.613 14.31 0.8934 9.099 8.781

    The variation of slope of fit lines correspondingto the strain gauges S2 and S3 clearly localized thedamage as shown in Figure-8. The plots are made onsimilar scale for easy comparison of the structuralcondition at the location of sensors. It is evident from

    Figure 8a that there is no significant change in the slope ofstrain gauge S1, which implies that no damage has takenplace within the coverage of the sensor. However,damages are clearly localized at the locations sensors S2and S3 as indicated by increase in slopes of fit lines in

    Figures 8b and 8c. Damages for scenario C2 is locatedclose to sensor S2, while damage scenario C3 is located inthe region of sensor S2 and S3. The statistical damageidentification method has reaffirmed that strain is verysensitive response to local damage. However, strain

    gauges cannot reflect influence of damage effectivelyunless they are located at the damaged region. Moreover,the challenges of huge data acquisition and processingcosts, durability and long-term reliability render straingauges unfit for large-scale civil SHM.

    45

  • 8/13/2019 Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

    9/10

    VOL. 4, NO. 3, MAY 2009 ISSN 1819-6608

    ARPN Journal of Engineering and Applied Sciences

    2006-2009 Asian Research Publishing Network (ARPN). All rights reserved.

    www.arpnjournals.com

    Figure-8.Variation of slopes of fit lines for strain gauges.

    Similar to strain gauge measurements, damages are evidently localized by FBG sensors F1 and F3 as shown in Figure-9.

    0

    20

    40

    60

    80

    0 5 10 15 20

    FRF magnitude of reference sensor S4

    FRF

    m

    agnitude

    ofS1

    C1S1 C2S1 C3S1

    0

    20

    40

    60

    80

    0 5 10 15 20

    FRF magnitude of reference sensor S4

    FRF

    m

    agnitu

    de

    ofS2

    C1S2 C2S2 C3S2

    0

    20

    40

    60

    80

    0 5 10 15 20

    FRF magnitude of reference sensor S4

    FRF

    ma

    gnitude

    ofS3

    C1S3 C2S3 C3S3

    0

    20

    40

    60

    80

    0 5 10 15 20 25MMS of Reference sensor F4

    M

    M

    So

    fFBGs

    ensorF1

    C1F1 C2F1 C3F1

    0

    20

    40

    60

    80

    0 5 10 15 20 25

    MMS of Reference sensor F4

    M

    M

    So

    fFBG

    sensorF2 C1F2 C2F2 C3F2

    0

    20

    40

    60

    80

    0 5 10 15 20 25

    MMS of reference sensor F4

    M

    M

    So

    fFBGs

    ensorF3

    C1F3 C2F3 C3F3

    Figure-9. Variation of slopes of fit lines for FBG sensors.

    There is no appreciable change in slope of sensorF1 in Figure-9a while Figures 9b and 9c show indicationsof damage in the regions covered by sensors F2 and F3.Damages for scenarios C2 and C3 are located within the

    gauge length of sensor F2 and damage scenario C3 islocated within sensor F3. One of the benefits of distributedstrain sensing technique over the conventional strain gaugeis that every critical section of the structure is covered

    without any loss of vital information for damageidentification.

    CONCLUSIONS

    A damage identification strategy adaptable tocivil structures using statistical features of dynamic strainmeasurements has been presented and verified throughexperimental study. The method has the capability to

    46

  • 8/13/2019 Adewuyi a., y z. Wu (2009) Vibration-based Structural Health Monitoring

    10/10

    VOL. 4, NO. 3, MAY 2009 ISSN 1819-6608

    ARPN Journal of Engineering and Applied Sciences

    2006-2009 Asian Research Publishing Network (ARPN). All rights reserved.

    www.arpnjournals.com

    reliably assess the consistency of the measurement data,promptly identify faulty sensors and accurately locatedamages in structure by using statistical features. The

    importance of the technique for civil SHM was establishedby using both point and distributed strain data andpresented in an easy-to-interpret graphical format foreffective implementation. The stability of all measurementdata was perfectly determined via the correlation of modalparameters of target sensor with the reference. The abilityof strain gauges to correctly locate damage is limited to itsshort gage length, while the long-gage FBG sensors aremore efficient choice for effective identification andlocalization of damage.

    REFERENCES

    Brincker R., P. Anderson, P.H. Kirkegaard and J.P.Ulfkjaer. 1995. Damage Detection in Laboratory ConcreteBeams. Proceedings of the 13th International ModalAnalysis Conference. pp. 668-674.

    Carden P.E. and P. Fanning. 2004. Vibration BasedCondition Monitoring: A Review. Structural HealthMonitoring. 3(4): 355-377.

    Doebling S.W. and C.R. Farrar. 1997. Using StatisticalAnalysis to Enhance Modal-Based Damage Identification.Structural Damage Assessment Using Advanced SignalProcessing Procedures. Proceedings of DAMAS 97

    University of Sheffield, UK. pp. 199-210.

    Doebling S.W., C.R. Farrar, M. B. Prime and D. W.Shevita. 1996. Damage identification and healthmonitoring of structural and mechanical systems fromchanges in their vibration characteristics: A literaturereview. Los Alamos National Laboratory Report.

    Farrar C.R., T.A. Duffey, S.W. Doebling and D.A. Nix.1999. A Statistical Pattern Recognition Paradigm forVibration-Based Structural Health Monitoring.Proceedings of the 2nd International Workshop onStructural Health Monitoring, Stanford, CA. Sept 8-10.

    Li S.Z. and Z.S. Wu. 2007. Development of DistributedLong-gage Fiber Optic Sensing System for StructuralHealth Monitoring. Structural Health Monitoring. 6(6):133-143.

    Li S.Z. and Z.S. Wu. 2008. Modal Analysis on Macro-strain Measurements from Distributed Long-gage FiberOptic Sensors. Journal of Intelligent Material Systems andStructures. 19(8): 937-946.

    Pape D.A. 1993. A Modal Analysis Approach to FlawDetection in Ceramic Insulators. Proceedings of the 11th

    International Modal Analysis Conference. pp. 35-40.

    Sohn H., C.R. Farrar, F.M. Hemez, D.D. Shunk, D.W.Stinemates and B. R. Nadler. 2003. A Review of

    Structural Health Monitoring Literature: 1996-2001. LosAlamos National Laboratory Report. LA-13976-MS.

    Wu Z. S. and S. Li. 2007. Two-level Damage DetectionStrategy Based on Modal Parameters from DistributedDynamic Macro-strain Measurements. Journal ofIntelligent Material Systems and Structures. 18(7): 667-676.

    47