sm2210

8/17/2019 sm2210

1/11

This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 203.110.242.7This content was downloaded on 09/05/2016 at 13:53

Please note that terms and conditions apply.

Damage detection in composite materials using Lamb wave methods

View the table of contents for this issue , or go to the journal homepage for more

2002 Smart Mater. Struct. 11 269

(http://iopscience.iop.org/0964-1726/11/2/310)

ome Search Collections Journals About Contact us My IOPscience

http://localhost/var/www/apps/conversion/tmp/scratch_1/iopscience.iop.org/page/termshttp://iopscience.iop.org/0964-1726/11/2http://iopscience.iop.org/0964-1726http://iopscience.iop.org/http://iopscience.iop.org/searchhttp://iopscience.iop.org/collectionshttp://iopscience.iop.org/journalshttp://iopscience.iop.org/page/aboutioppublishinghttp://iopscience.iop.org/contacthttp://iopscience.iop.org/myiopsciencehttp://iopscience.iop.org/myiopsciencehttp://iopscience.iop.org/contacthttp://iopscience.iop.org/page/aboutioppublishinghttp://iopscience.iop.org/journalshttp://iopscience.iop.org/collectionshttp://iopscience.iop.org/searchhttp://iopscience.iop.org/http://iopscience.iop.org/0964-1726http://iopscience.iop.org/0964-1726/11/2http://localhost/var/www/apps/conversion/tmp/scratch_1/iopscience.iop.org/page/terms

8/17/2019 sm2210

2/11

INSTITUTE OF PHYSICS PUBLISHING SMART MATERIALS AND STRUCTURES

Smart Mater. Struct. 11 (2002) 269–278 PII: S0964-1726(02)33619-X

Damage detection in composite materialsusing Lamb wave methodsSeth S Kessler 1,3, S Mark Spearing 1 and Constantinos Soutis 2

1 Department of Aeronautics and Astronautics, Massachusetts Institute of Technology,Cambridge, MA 02139, USA2 Department of Aeronautics, Imperial College of Science, Technology and Medicine,London SW7 2BY, UK

E-mail: [email protected]

Received 6 August 2001, in nal form 24 January 2002Published 5 April 2002

Online at stacks.iop.org/SMS/11/269AbstractCost-effective and reliable damage detection is critical for the utilization of composite materials. This paper presents part of an experimental andanalytical survey of candidate methods for in situ damage detection of composite materials. Experimental results are presented for the applicationof Lamb wave techniques to quasi-isotropic graphite/epoxy test specimenscontaining representative damage modes, including delamination, transverseply cracks and through-holes. Linear wave scans were performed on narrowlaminated specimens and sandwich beams with various cores by monitoringthe transmitted waves with piezoceramic sensors. Optimal actuator andsensor congurations were devised through experimentation, and varioustypes of driving signal were explored. These experiments provided aprocedure capable of easily and accurately determining the time of ight of a Lamb wave pulse between an actuator and sensor. Lamb wave techniquesprovide more information about damage presence and severity thanpreviously tested methods (frequency response techniques), and provide thepossibility of determining damage location due to their local responsenature. These methods may prove suitable for structural health monitoringapplications since they travel long distances and can be applied withconformable piezoelectric actuators and sensors that require little power.

(Some gures in this article are in colour only in the electronic version)

1. Introduction

1.1. Health monitoring of composite structures

Structural health monitoring (SHM) has been dened inthe literature as the ‘acquisition, validation and analysisof technical data to facilitate life-cycle managementdecisions’ [1]. More generally, SHMdenotesa systemwith theability to detect and interpret adverse ‘changes’ in a structurein order to improve reliability and reduce life-cycle costs. Thegreatest challenge in designing an SHM system is knowingwhat ‘changes’ to look for and how to identify them. Thecharacteristics of damage in a particular structure play a keyrole in dening the architecture of the SHM system. The

3 Author to whom any correspondence should be addressed.

resulting ‘changes’, or damage signature, will dictate thetype of sensor that is required, which in turn determines therequirements for the rest of thecomponents in the system. Thepresent research project focuses on the relationship betweenvarious sensors and their ability to detect ‘changes’ in astructure’s behavior.

The aerospace industry has one of the highest payoffs forSHM since damage can lead to catastrophic (and expensive)failures, and the vehicles involved have regular costlyinspections. Currently 27% of an average aircraft’s life cyclecost is spent on inspection and repair [2]: a gure thatexcludes the opportunity cost associated with the time theaircraft is grounded. These commercial and military vehiclesare increasingly using composite materials to take advantageof their excellent specic strength and stiffness properties,

0964-1726/02/020269+10$30.00 © 2002 IOP Publishing Ltd Printed in the UK 269

http://stacks.iop.org/sm/11/269

8/17/2019 sm2210

3/11

8/17/2019 sm2210

4/11


of a Lamb wave in a particular material is with their dispersioncurves, which plot the phase and group velocities versus theexcitation frequency. The derivation of these curves beginswith the solution to the wave equation for the anti-symmetricLamb wave as seen in equation (1)

tan d̄

1 − ζ 2

tan d̄ ξ 2 − ζ 2+

(2ζ 2 − 1)2

4ζ 2 1 − ζ 2 ξ 2− ζ 2= 0 (1)

where the non-dimensional parameters are

ξ 2 =c2t c2l

, ζ 2 =c2t

c2phase, d̄ =

kt t 2

. (2)

These velocities can be dened by Lam é’s constants:

µ =E

2(1 + ν), λ =

Eν(1 − 2ν)( 1 + ν)

(3)

c2t =

µρ , c

2l =

(λ + 2µ)ρ , k t =

ωct . (4)

Substituting these equalities into the non-dimensionalparameters yields

ξ 2 =µ

(λ + 2µ)=

1 − 2ν2 − 2ν

,

ζ 2 =µ

ρc 2phase=

E

2ρ( 1 + ν)c 2phase

d̄ =ωt 2ct

=ωt 2 ρµ = ωt 2 2ρ( 1 + ν)E .

(5)

Finally, equation (5) can be substituted into equation (1) tobe solved numerically. For a given material, the Young’smodulus E , Poisson ratio ν and the density ρ are known, andthephase velocity cphase is thedependentvariablebeing solved.The independent variable being iteratively solved for is thefrequency–thickness product, where ω is thedriving frequencyin rad s− 1. An example of a phase velocity dispersion curvefor the rst anti-symmetric Lamb wave using the materialproperties from the specimens used in the present research canbe seen in gure 2. The other useful plot is the group velocitydispersion curve, which can easily be derived from the phasevelocity curve using equation (6):

k = 2πλ w

(wave #) , λ w =cphase

f (wavelength )

cgroup = cphase + ∂c phase

∂kk =

cphase1 − (f/c phase )∂c phase /∂f

.

(6)An example of a group velocity dispersion curve, again usingthe material properties from the present research, can be seenin gure 3. The equations presented here are intended forisotropicmaterials; however, it hasbeen shown in the literaturethat A0 is fairly invariant to the layup of a composite material,and can be closely approximated by using the bulk laminateproperties [13]. Finite-element techniques have been used byother researchers in the literature to more accurately determinethe wave velocities in composite materials [12,15].

The group velocity of a Lamb wave produced by apiezoelectric patch driven at a particular frequency can easily

Figure 1. Graphical representation of A and S Lamb wave shapes.

Figure 2. Phase velocity dispersion curve for the A 0 mode of aneight-ply composite laminate.

Figure 3. Group velocity dispersion curve for the A 0 mode of aneight-ply composite laminate.

be veried in a control (undamaged) specimen by measuringthe TOF in an oscilloscope between two sensors of knownseparation. This information can then be used to locatedamaged areas along a specimen, without using any analyticalmodels, by observing the disturbed wave between the sensorand actuator. The Lamb wave’s group velocity essentiallyvaries by a similar equality to that of a structure’s resonantfrequency, as (E/ρ) 1/ 2, where E is the modulus and ρ isdensity, so as a wave travels across an area of reduced stiffnessit will slow down. Another phenomenon associated withdamage is analogous to traveling acoustical waves; uponreaching a region of dissimilar wave speed, a portion of thewave is reected proportionally to the difference in their

271

8/17/2019 sm2210

5/11

S S Kessler et al

Figure 4. CFRP specimen (250 mm × 50 mm) with piezoceramicactuator and sensors.

stiffness and density. From these two pieces of information,good correlation with damage location and magnitude canbe determined. Several experiments were performed in thepresent research to establish theeffectiveness of this procedurein determining the damage present in a composite laminate.

3. Optimization

There is currentlyno standard or even a best-practice precedentfor damage detection via Lamb wave testing. Severalprocedures have been developed in the literature, each withvaluable characteristics, and each with some degree of arbitrariness. The goal of the rst part of the present researchwas to determine experimentally and analytically what effectsvariousparametershave on the sensitivity of damage detection.These parameterscanbe dividedinto threecategories: actuatorand sensor geometry, actuation pulse and specimen properties.Therst of these categorieswas notexplored thoroughly in thisresearch. A few shapes of piezoceramic patches were usedto produce Lamb waves, and as expected waves propagatedparallel to each edge, i.e. longitudinally and transversely for a

rectangular patch and circumferentially from a circular piezo.Several other researchers have examined the effects of piezodimensions on their actuation, thus it was not of particularinterest for the present research [28,29].

The second set of variables explored was the actuationpulse parameters. These included the pulse shape, amplitude,frequency and number of cycles to be sent during each pulseperiod. These parameters were varied experimentally on acontrol specimen to observe their effect on the Lamb wavesgenerated. Two PZT piezoceramic patches were attached toeither end of the specimen, one connected to an arbitraryfunction generator and the other to an oscilloscope as seen

in gure 4. For each pulse shape, the various parameters werechanged independently as the transmitted wave was observedin the oscilloscope. The shapes that produced the best resultswere then compared inMatlab TM byusingtheir power-spectral-density (PSD) plots. Similarly the effect of the number of cycles perperiod for thedifferent-shaped signals was observedin the PSD plots by comparing the energy dedicated to theprincipal driving frequency. The more energy dedicated to thedesired driving frequency, the stronger the Lamb wave and themore accurate the wave speed calculation, and hence the moresensitive and reliable the damage detection capability.

The nal component of the optimization analysismathematically quantied the signicance of the specimens’geometric and material properties. The Lamb solution forthe wave equation was used to plot the dispersion curvesfor each proposed specimen conguration as described in theprevious section. The material constants for the composite

laminatestobe analyzedwerecalculated by classical laminatedplate theory, and then input into the mathematical model [30].Using the same model code, each material constant suchas the tensile modulus, Poisson ratio and density weremodied independently, and the effect on the dispersioncurves was documented. Since the dependent variable in

thismathematical modelwas the frequency–thicknessproduct,changes in specimen thickness were easily quantiable sincethey proportionally shifted the plot along the x-axis. From theoptimization experiments and analysis an effective testmethodwas determined, which was then used to detect damage inseveral test specimens.

4. Experimental procedure

4.1. Narrow coupon tests

Following a building block approach [31], the rstset of experiments was conducted on narrow compositecoupons. The laminates used for this present researchwere manufactured during previous research that exploredfrequency response methods (FRMs) as a means of damagedetection, and were re-used to compare directly theeffectiveness of the two methods [32]. The specimens were25 cm × 5 cm rectangular [90 / ± 45/ 0]s quasi-isotropiclaminates of the AS4/3501-6 graphite/epoxy system, whichwere clamped on one end to match the boundary conditionsfrom the previous research (however, experimentation provedthat the boundary conditions around the frame of the specimenhad no effect on the Lamb wave traveling between twopiezoceramic patches). Three PZT piezoceramic patches were

afxed to each specimen, as shown in gure 4, using 3MThermoBond TM thermoplastic tape so that they were rmlyattached during testing, but could be removed afterwards torecover the specimens for future tests. The PZT was cut into2 cm × 0.5 cm patches so that the longitudinal wave wouldbe favored over the transverse one, and three patches wereused on each specimen to actuate and accurately measure thetransmitted and reected waves. Both the actuation and thedataacquisition wereperformedusinga portable NI-Daqpad TM6070E data acquisition board, anda laptop running Labview TMas a virtual controller. A Labview TM VI-le was created whichwould load an arbitrary waveform from Matlab TM and output itat the desired frequency and amplitude, while simultaneouslyacquiringdata on four channels at 600000samples persecond.The rst channel, which served as the trigger for all of thechannels, was connected to the output channel and actuatingPZT, two others were connected to the sensing piezoceramicpatches andthe nal channel was connectedto a PZTsensornotattachedtothespecimentoserveasacontrolchannelinordertozero out drift. A single pulse of the optimal signal found in theprevious section, shown in gure 5, was sent to thedriving PZTpatch to stimulate an A 0 mode Lamb wave, and concurrentlythe strain-induced voltage outputs of the other two patcheswere recorded for 1 ms to monitor the wave propagation.

The resulting data were then passed to Matlab TM wherethedrift wasltered out and thewaveforms could be comparedand analyzed within two specialized toolboxes. In the signalprocessing toolbox the waves could be easily superimposed,and a built-in peak detector was used to determine accurately

272

8/17/2019 sm2210

6/11


Figure 5. Actuation signal used to generate A 0 Lamb mode, 3.5sine waves at 15 kHz.

the TOF for each signal, and the delay in time of arrivalbetween two specimens. Subsequently, in the wavelet toolboxa DB3 wavelet was used to decompose the data into itsfrequency components. By plotting the magnitude of thewavelet coefcient at the peak driving frequency, the energyremaining from the input signal could be compared [33]. Thisprocedure was carried out for two of each specimen type at thedriving frequency of 15 kHz.

As with the previous research on FRM [32], various typesof damage were introduced to the specimens. In the rstgroup, 6.4 mm diameter holes were drilled into the centerof each specimen as a stress concentration. The next groupwas compressively loaded in a four-point bending xture untilaudible ber fracture damage was heard, and the third wascyclically loaded in the same xture for 2000 cycles at 80% of this load with an R ratio of − 1 to create matrix cracks. Thenext two groups of specimens were delamination specimenswhich were introduced by two methods: one used a thinutility blade to cut a 5 cm × 2.5 cm slot in one side, andthe other with a Teon strip cured into the center mid-planeof the laminate. After the damage was introduced into eachspecimen, an x-ray radiograph was taken using a die-penetrantto help document the type, degree and location of the damage

as shown in gure 6.

4.2. Sandwich coupon tests

Analogous experiments were performed on sandwich couponsto that of the narrow laminates in order to test the effect of various types of core material on the propagation of Lambwaves. Four different cores were used: low- and high-density(referred to as LD and HD) aluminum honeycomb, Nomex TMand Rohacell TM . Each specimen contained two facesheetsidentical to the undamaged laminates in the previous sectionsurrounding a 2 cm thick core, which were adhered using FM-123 lm adhesive in a secondary curing process. Two controlsand two damaged specimens of each type were manufacturedfor testing. In the damaged specimens, a 5 cm × 2.5 cm pieceof Teon was placed between the adhesive and the core ina central 2.5 cm region during the cure so that the facesheet

Figure 6. X-radiographs of damaged specimens. ( a ) Controlspecimen with no damage present. ( b) Stress concentrationspecimen with drilled through-hole. ( c) Matrix-crack specimen withfatigue-induced damage. ( d ) Delamination specimen cut with a thinutility knife at the mid-plane.

would not bond to the honeycomb to simulate a delamination.An additional specimen was also manufactured with the high-density aluminum core that had a 2 cm diameter circular pieceof Teon placed between the layers on either side so that it wasindistinguishable from the controls by sight. This specimenwas used for a ‘blind test’ of the proposed Lamb wave damagedetectionmethod, where it was testedalongside thetwo controlspecimens to determine which had the articial aw. Thetest setup and data analysis procedure for the sandwich beamexperiments were identical to those of the thin specimens withthe exception of the driving frequency, which was determinedto be more effective at 50 kHz for these tests.

5. Results

5.1. Optimization results

5.1.1. Pulse frequency. The dispersion curves show therelationship between the phase velocity and pulse frequency.At lower frequencies, fewer Lamb modes are excited so theresponse signal is more distinguishable, and the velocity isslower so there is more time separating the sent and receivedsignals, making any changes more distinguishable. At theselower frequencies however, the dispersion curves have steepslopes and thus are very sensitive to small variations infrequency making it difcult to predict the TOF. At higherfrequencies,when more modesarepresent, theslopethen tendsto atten out with the consequence of a shorter wave pulsecarrying less or more compressed information on the damage.For the experiments performed during this present research,15 kHz was chosen for the thin laminate tests, and 50 kHz forthe sandwich beam tests as the optimal testing points. Thesefrequencies were based on their slope and location on thedispersion curves, evidence from previous research suggestingthese frequencies for specimens of similar geometries, andbrief experimentation using a function generator to determinethe maximum response amplitude for a range of driving

frequencies.

5.1.2. Pulse amplitude. The increase of the drivingvoltage proportionately increases the magnitude of the

273

8/17/2019 sm2210

7/11

S S Kessler et al

Lamb wave strain. In these experiments, driving thepiezopatches at an amplitude of 5–10 V produced a 10–25 mVresponse due to the wave sensed by the PZT patch. Increasingthe amplitude also increases the signal-to-noise ratio to yield aclearer signal, since the static noise received by the PZT patchis usually in the 1–5 mV range. Higher voltage however also

tended to increase thedrift in the signal, which deteriorated theresolution capabilities of the data acquisition system. Also, apotential SHM system should be as low power as possible.The optimal driving voltage was therefore chosen to be 5 V forthese experiments.

5.1.3. Number of cycles. The number of cycles of a periodicfunction with which to actuate the piezopatches is one of the more complicated decisions to be made for Lamb wavetechniques. The FFT of a continuous sine wave would yield asingle peak at the driving frequency; however, for a few nitecycles, the FFT appears as a Gaussian curve with a peak at

the driving frequency. Thus, the more periods of a wave sentinto a driving pulse, the narrower the bandwidth and the lessdispersion. The problem in a short specimen though, is thegreater the number of periods of a wave in the pulse, the lesstime between the last sent signal and the rst reected one,so the response is more difcult to interpret. An appropriatenumber of cycles can be determined by the maximum numberof waves that can be sent in the time it takes for the lead waveto travel to the sensing PZT patch. It is also convenient to useintervalsof half cycles so that thesentsinusoidalpulsebecomessymmetric. Research from the literature has used signalsvarying from 3.5 to 13.5 cycles per actuating pulse [17–24].Since the specimens in thecurrent research are relatively short,few cycles could be actuated without disturbing the receivedsignal, thus 3.5 cycles were used to drive the piezoceramicactuators.

5.1.4. Pulse shape. Of the signal shapes that wereanalyzed and experimented on, pure sinusoidal shapes appearto excite Lamb wave harmonics most efciently, since theyare periodic, smooth and have comparatively quick risetimes to their peak amplitude as compared with a parabolicshape. A Hanning window (approximated by a half-sinewave multiplied over the pulse width) helps to narrow thebandwidth further to focus the maximum amount of energyinto the desired actuating frequency with the least ‘spill-over’from neighboring frequencies.

5.1.5. Material properties. The relationship between thematerial properties of a specimen and the speed of thepropagating Lamb wave is quite complex; however, anunderstanding is necessary to design an appropriate damagedetection test. To rst order, the wave velocity increases withthe square root of the modulus, i.e. an increase in modulusslightly speeds the wave velocity. An increase in the densitywould have the opposite effect however by slowing wavevelocity, as it appears in all the same terms as the modulus buton the reciprocal side of the divisor. The effect of the Poissonratio is probably the most complicated, as it appears in mostterms; however, to rst order, small changes seem to have littleto no effect on the wave velocity. The most straightforward

parameter is the thickness of the specimen, which has a linearrelationship with the Lamb wave velocity. The thicker thespecimen the quicker the speed and the higher the dispersionrate for a given driving frequency.

5.2. Experimental testing results

There were two sets of results obtained for both the thincoupons and the narrow sandwich beams. The rst set of results included the virgin time traces of voltage from thePZT sensor at the far end of the specimen. For the thincoupons, 1 ms of data were taken and the average peak voltagewas around 20 mV. The time traces for one of each type of specimen along with a superimposed control specimen areshown in gure 7. Similarly, 500 µ s of data were taken forthe sandwich beams with an average peak voltage of around10 mV. For these specimens, time traces of each control beamare plotted against their delaminated complement in gure 8.In each of these plots, a ‘bleed-through’ portion of the sentsignal leaking across the data acquisition board can be seenat the beginning of the time trace. Since the channels wereall triggered at the 5 V peak voltage, exactly half of the sentsignal is visible so this became a convenient way to measuretheTOF. Thesecond setof results for each specimen group wasthe outcome of thewaveletdecomposition. Foreachspecimen,the ‘bleed-through’ portion of the signal was ltered out, andthe wavelet coefcient magnitude of the dominant frequency(15 kHz for the thin coupons and 50 kHz for the beams) wasplotted over time. For the thin coupons, gure 9 comparesthese coefcients, and thus the transmitted energy, for oneof each type of specimen. Finally, gure 10 displays the

coefcient magnitude results for the ‘blind test’, comparingthe two high-density aluminum core control specimens withone known and one unknown damaged specimen.

6. Discussion

6.1. Experimental procedure optimization

There were two important sets of guidelines obtained from theoptimization portion of this research. The rst set was theanalytical trade studies performed to predict the effectivenessof Lamb wave methods in different applications. Using themathematical formulations derived by Lamb, the effects of material constants and specimen geometry were determined.By entering the material properties for a particular applicationthe resulting dispersion curves provide a range of potentialwavevelocities for theA 0 modedriven at differentfrequencies.If the characteristic wave velocity for a material is too fast tobe reasonably acquired, then this method is not suitable. Also,with knowledge of the effects of various damage types on thestiffness of a particular material, the resolution of change forthe resultant signal, or ‘observeability’ can be predicted inorder to determine thedetection limitations with respect to awsize for a given data acquisition capability. This procedurewasconducted prior to the experimental procedure of this research,and it was determined that graphite/epoxy composite materialwas a good candidate for Lamb wave methods, and that withthe detection capabilities of the data acquisition system areasonable change in stiffness (5–10%) could be resolved.

274

8/17/2019 sm2210

8/11


Figure 7. Time trace of voltage signal from PZT sensor 20 cm from actuator, 15 kHz signal. Solid curves are damaged specimens; control issuperimposed as a dashed curve.

Figure 8. Time trace of voltage signal from PZT sensor 20 cm from actuator, 50 kHz signal. Solid curves are undamaged beam controls;debonded specimens have dashed curves.

The second guideline provided an experimental andanalytical determination of the optimal testing congurationto be applied for a particular material, including the drivingfrequency, numberof cycles and theminimal required distancefor placement of the actuators away from boundaries orfeatures. Using the dispersion curves for a material, arange of potential driving frequencies can be selected based

upon regions of smallest slope and driving capabilities whileremaining below frequencies that would generate higher-orderwaves. Next, experimentally these frequencies can be tunedusing a function generator to nd the optimal frequency in that

range that produces the largest-amplitude Lamb wave. Thereis then a trade between the number of waves that can be sent ina pulse and the distance from abrupt features in the structure;as discussed previously the more waves the more energy thatgoes into the Lamb wave; however, if features are too close tothe actuator/sensor with many waves then changes in responsesignal may be obscured. For the experiments presented in

this paper, it was determined that 15 kHz was the optimaldriving frequency for the thin laminates and 50 kHz for thesandwich beam specimen, using a signal of 3.5 sine wavessince the specimen was relatively short. These sets of tools

275

8/17/2019 sm2210

9/11

S S Kessler et al

Figure 9. Wavelet coefcient plots for thin coupons; compares 15 kHz energy content.

Figure 10. Wavelet coefcient plots for beam ‘blind test’; compares 50 kHz energy content.

could be used in tandem by an engineer developing an SHMsystem for a vehicle to decide if the Lamb wave method wouldprovidesatisfyingresults for theirapplication, andto determinethe appropriate driving parameters to obtain the best damagedetection resolution.

6.2. Interpretation of experimental data

There are generally ve goals for damage detection, each of which is gained with increasing difculty and complexity.The rst is the determination of the presence of damage in aspecimen. Thesecond is an estimation of theextent of severity

of the damage. The third goal is to be able to differentiatebetween various different types of damage. The fourth is to beable to calculate where the damage is located. The nal goalis to estimate the size of the damage. It appears that Lambwave methods carry enough information potentially to meetall of these goals with a strategically placed array of sensorsand suitable processing codes; however, the current scope of this research focuses on the rst two goals.

The results from the narrow coupon tests clearly show thepresence of damage in all of the specimens. First of all, whenthe time traces of all of the control specimens were overlaid,there was a high degree of visible correlation, especially for

276

8/17/2019 sm2210

10/11


the rst half of the voltage time trace. The slight variation inthe second half of the data can be attributed to the reectedsignals returning from the far end of the specimen and passingunder the PZT sensor again, which may encounter a slightcutting bias in the composite to cause a change in phase.Of the articially damaged specimens, the Teon-induced

delamination was most easily quantied. When comparedwith the control specimens, these time traces appear at thesame phase and frequency, only having been delayed about55 µ s due to the damage. For the other types of damagethe frequency often remained the same; however, there wasa large reduction in amplitude, and a large and varying changein phase. This time trace was reproducible within a singlespecimen, although it would not be consistent across multiplespecimens with identical forms of damage. This is due tothe scatter and reecting of the waves on the various featuresof damage which may not be identical specimen to specimen,which makesa ‘damage signature’ difcult to dene. Themost

distinctly altered signal was that of the through-hole, havingthe same diameter as the actuator and sensor widths, whichhad the smallest voltage magnitude of all the specimens. Themost obvious method to distinguish between damaged andundamaged specimens however is by regarding the waveletdecomposition plots. The control specimens retained overtwice as much energy at the peak frequency as compared withall of the damaged specimens, and especially contained muchmore energy in the reected waves. The loss of energy inthe damaged specimens again is due to the dispersion causedby the micro-cracks within the laminate in the excitation of high-frequency local modes.

Thesandwichbeamresultsweremoredifcult to interpret,due to the damping nature of the cores signicantly reducingthe voltage generated by the PZT sensors. The high-densityaluminum core, which was the stiffest of the four tested,provided the clearest results; the other specimens yieldeddecreasing magnitude voltages as the stiffness decreased thusincreasing the damping factor. There were two basic trendsacross all the specimens. The rst was that the responsesof the control specimens were larger than those that weredelaminated for each core type. This is most likely dueto the loss of energy of the wave in a local mode over thedelaminated region. The second trend was the appearance of more reected waves after the initial pulse in the time tracein the delaminated specimens, which again was probably dueto other higher-frequency modes being excited in the regionof reduced thickness and dampening. Probably the mostsignicant result of the present research was the ‘blind test’.Four high-density aluminum beam specimens were tested, oneof which had a known delamination in its center, while of theremaining three specimens it was unknown which containedthe circular disbond and which two were the undamagedcontrols. By comparing the four wavelet coefcient plots ingure 10, onecaneasilydeduce that the two control specimensare the ones with much more energy in the transmitted signals,while the third specimen (control C) obviously has the awthat reduces energy to a similar level to that of the knowndelaminated specimen. This test serves as a true testament tothe viability of the Lamb wave method being able to detectdamage in at least simple structures.

6.3. Implementation of Lamb wave techniques in an SHM system

Lambwave techniqueshave goodpotential for implementationin an SHMsystem. These methods provide useful informationabout the presence, location, type, size and extent of damagein composite materials, and can be applied to a structure withconformable piezoelectric devices. The major disadvantageof this method is that it is active; it requires a voltagesupply and function generating signal to be supplied. Thiscan be complicated in a large structure, especially if theSHM system is to be implemented wirelessly; it has beensuggested in the literature however that PZT can be actuatedremotely using radio-frequency waves [24]. Another difcultrequirement is the high data acquisition rate needed to gainuseful signal resolution. If a system is sampling at 0.5 MHzfrom several sensors, a large volume of data will accumulatequickly; this implies the need for local processing. The dataacquisition capabilities dictate the limitations of aw size

able to be resolved by a system using this method. In orderto conserve power and data storage space, the Lamb wavemethod should most likely be placed into an SHM systemin conjunction with another passive detection method, suchas an FRM. The piezoelectric patches used to actuate theLamb waves could passively record frequency response datauntil a certain threshold of change is surpassed, and thentrigger the generation of Lamb waves to gain more specicdata about the damaged region. Three to four piezoelectricmulti-functioning actuator/sensor patches would be placed inthe same vicinity in order to be able to triangulate damagelocation based upon reciprocal times of ight and reected

waves. The separation between sensing patches would dependon several parameters such as thematerial properties, dampingcharacteristics and curvature of the structure, and for at areascould be as large as 2 m apart [24]. The detailed specicationsof the Lamb wave method to be used for a particularapplication would be designed by the procedure described inthe previous optimization section. Another useful detectioncapability arises fromthe fact that twodifferent optimal drivingfrequencies were necessary for the thin laminates and thebeam structures. This offers the possibility of having theability to differentiate between damage within the laminateversus damagebetween the laminate and the core by discretelydriving at two different frequencies. This procedure was notexplored during the present research; however, preliminaryexperimentation indicates that the potential of this procedureworking exists.

7. Conclusions

This paper has explored the optimization and applicationof Lamb wave methods to damage detection in compositematerials. Several mathematical trade studies were performedto observe the effect of various material constants andactuator driving parameters. Using these tools, an optimalconguration was selected for the experimental section of this research. For this optimal procedure, several narrowgraphite/epoxy specimens were tested with various forms of pre-existing damage, such as delamination, matrix cracks andthrough-thickness holes. Similar tests were also performed

277

8/17/2019 sm2210

11/11

S S Kessler et al

on narrow sandwich beams using cores of various densitiesand stiffness. These tests demonstrated the feasibility of detecting several types of aw in representative compositestructures, and this method was validated successfully by a‘blind test’ of several beam specimens. Lamb wave techniqueshave the shown the potential to provide more information

than previously tested methods such as FRM since they aremore sensitive to the local effects of damage in a material ascompared with the global nature of FRM [32]. Similar tofrequency response methods, their results are limited at higherfrequencies; however, their low-frequency results shouldprovide sufcient data to predict damage. The disadvantageof Lamb wave methods is that they require an active drivingmechanism to propagate the waves, and the resulting datacan be more complicated to interpret than for many othertechniques. Overall however, Lamb wave methods have beenfound to be effective for the in situ determination of thepresence and severity of damage in composite materials, andhold the potential to locate damage due to their local responsenature, which will be addressed in future research. Furtherexperimentation will be aimed at testing two-dimensional andbuilt-up structures using this technique, and the applicationof Lamb wave methods using a single multi-purpose actuatorand sensor. SHM systems will be an important component infuture designs of air and spacecraft to increase the feasibilityof their missions, and Lamb wave techniques will likely playa role in these systems.

References

[1] Hall S R 1999 The effective management and use of structural

health data Proc. 2nd Int. Workshop on Structural Health Monitoring pp 265–75[2] Hall S R and Conquest T J 1999 The total data integrity

initiative—structural health monitoring, the next generationProc. USAF ASIP Conf. 2nd edn

[3] Bhat N 1993 Delamination Growth in Graphite/EpoxyComposite Laminates under Tensile Load (Cambridge, MA:Massachusetts Institute of Technology)

[4] Wang S S 1979 Delamination crack growth in unidirectionalber-reinforced laminates under static and cyclic loadingComposite Material Testing and Design ASTM STP (674)pp 642–63

[5] Bar-Cohen Y 1986 NDE of ber reinforced compositematerials—a review Mater. Eval. 44 446–54

[6] Chang F K 1999 Structural health monitoring: a summary

report Proc. 2nd Int. Workshop on Structural Health Monitoring (Stanford, CA, 1999)[7] Giurgiutiu V, Jingjing B and Zhao W 2001 Active sensor wave

propagation health monitoring of bean and plate structuresProc. SPIE Int. Symp. on Smart Structures and Material

[8] Worlton D C 1961 Experimental conrmation of Lamb wavesat megacycle frequencies J. Appl. Phys. 32 967–71

[9] Demer L J and Fentnor L H 1969 Lamb wave techniques innondestructive testing Int. J. Nondestruct. Test. 1 251–83

[10] Saravanos D A and Heyliger P R 1995 Coupled layerwiseanalysis of composite beans with embedded piezoelectricsensors and actuators J. Int. Mater. Syst. Struct. 6 350–62

[11] Saravanos D A, Birman V and Hopkins D A 1994 Detection of delaminations in composite beams using piezoelectricsensors Proc. 35th Structures, Structural Dynamics and

Materials Conf. (AIAA)

[12] Percival W J and Birt E A 1997 A study of Lamb wavepropagation in carbon-bre composites Insight, Non-Destr.Test. Cond. Monit. 39 728–35

[13] Birt E A 1998 Damage detection in carbon-bre compositesusing ultrasonic Lamb waves Insight, Non-Destr. Test.Cond. Monit. 40 335–9

[14] Badcock R A and Birt E A 2000 The use of 0–3piezocomposite embedded Lamb wave sensors for detectionof damage in advanced bre composites Smart Mater.Struct. 9 291–7

[15] Seale M D, Smith B T and Prosser W H 1998 Lamb waveassessment of fatigue and thermal damage in composites J. Acoust. Soc. Am. 103 2416–24

[16] Tang B and Henneke E G 1989 Lamb wave monitoring of axial stiffness reduction of laminated composite plates Mater. Eval. 47 928–33.

[17] Rose J L, Hay T and Agarwala V S 2000 Skin to honeycombcore delamination detection with guided waves 4th Joint DOD/FAA/NASA Conf. on Aging Aircraft

[18] Osmont D, Devillers D and Taillade F 2000 A piezoelectricbased health monitoring system for sandwich platessubmitted to damaging impacts Eur. Congr. onComputational Methods in Applied Sciences and Engineering

[19] Monkhouse R S C, Wilcox P D and Cawley P 1997 Flexibleinterdigital PVDF transducers for the generation of Lambwaves in structures Ultrasonics 35 489–98

[20] Monkhouse R S C, Wilcox P W, Lowe M J S, Dalton R P andCawley P 2000 The rapid monitoring of structures usinginterdigital Lamb wave transducers Smart Mater. Struct. 9304–9

[21] Valdez S H D and Soutis C 2000 Structural integritymonitoring of CFRP laminates using piezoelectric devicesProc. Eur. Conf. on Composite Materials

[22] Valdez S H D and Soutis C 2000 Health monitoring of composites using Lamb waves generated by piezoelectricdevices Plast. Rubber Compos. 29 475–81

[23] Valdez S H D and Soutis C 2001 A structural healthmonitoring system for laminated composites Proc. 18th Biennial Conf. on Mechanical Vibration and Noise

[24] Valdez S H D September 2000 Structural integrity monitoringof CFRP laminates using piezoelectric devices PhD ThesisImperial College of Science, Technology and Medicine

[25] Viktorov I A 1967 Rayleigh and Lamb Waves, Physical Theoryand Applications (New York: Plenum)

[26] Nayfeh A H 1995 Wave Propagation in Layered Anisotropic Media With Applications to Composites vol 39(Amsterdam: Elsevier)

[27] Lamb H 1917 On waves in an elastic plate Proc. R. Soc. A 93293–312

[28] Fripp M 1995 Distributed structural actuation and control withelectrostrictors SM Thesis Massachusetts Institute of

Technology[29] Lively P 2000 Dynamic structural shape estimation usingintegral sensor arrays SM Thesis Massachusetts Institute of Technology

[30] Jones R M 1999 Mechanics of Composite Materials 2nd edn(Blacksburg, VA: Taylor and Francis)

[31] The Composite Materials Handbook MIL-17 1999 Guidelines for Characterization of Structural Materials vol 1MIL-HDBK-1E (Department of Defense)

[32] Kessler S S, Spearing S M, Atalla M J, Cesnik C E S andSoutis C 2001 Structural health monitoring in compositematerials using frequency response methods Proc. SPIE Int.Symp. on Smart Structures and Material

[33] Lind R, Kyle S and Brenner M 2001 Wavelet analysis tocharacterize non-linearities and predict limit cycles of an

aeroelastic system Mech. Syst. Signal Process. 15 337–56

278

sm2210

Documents