1

Detection of multiple low-energy impact damage in composites plates using Lamb wave technique

Pedro André Viegas Ochôa de Carvalho a, b

a Department of Mechanical Engineering, Instituto Superior Técnico, Avenida Rovisco Pais, 1, 1049-001 Lisboa, Portugal

b Optical Non-Destructive Testing Laboratory, Faculty of Aerospace Engineering, Delft University of Technology, Kluyverweg 1, 2629 HS Delft, The Netherlands

Abstract

This work concerns the assessment of the suitability of the Lamb wave method, in particular of the two zero-order Lamb modes (A0 and S0), to detect multiple barely-visible impact damage in composite material. Four plates were produced with carbon-epoxy cured pre-preg, using a representative stacking sequence. Three specimens were subjected to multiple impact damage at three different low-energy levels, and one was left as an undamaged reference sample. Ultrasonic Lamb wave modes were selectively generated by surface-bonded piezoceramic wafer transducers in two tuned configurations. A signal identification algorithm in the time-scale domain based on the Akaike Information Criterion (AIC) was used to determine the group velocity of the Lamb modes.

The effectiveness of the Lamb wave method was successfully verified on all damage scenarios, since the 5 and 10 J damages were undoubtedly detected by the S0 mode configuration. The results were validated by digital Shearography, ultrasonic C-scan, and optical microscope observations, revealing strong consistency. For the material tested, the detection threshold of the three NDT methods was found to be between 3 and 5 J. Keywords: Composite materials, Non-Destructive Testing (NDT), Lamb wave, multiple barely-visible impact damage, piezoceramic transducers.

1. Introduction

There is a growing concern in the aircraft industry to increase the ratio of the structure effectiveness to the acquisition and utilization costs, which is called structure cost-effectiveness [1]. Assuming the acquisition costs are fixed, one of the major steps towards the reduction of utilization costs and the increase of structure effectiveness has been the widespread use of composite materials. Their excellent strength-to-weight ratio maximizes the structure capability. Their corrosion and fatigue resistance increases the time to failure, increasing reliability and reducing maintenance costs. Furthermore, their low density allows lower fuel consumption, and therefore lower operation costs.

However, contrary to metallic materials, one of the most serious issues related to the use of composites in airframes is their brittle behaviour in the presence of barely-visible impact damage (BVID), which may lead to unexpected failure under fatigue loading [2]. Therefore, the Non-Destructive Testing (NDT) techniques, that have already proven to be able to enhance safety, integrity and durability of aircraft structures over the last fifty years, combined with the recently developed measuring and computational technologies, assume a central role in the implementation of Structural Health Monitoring (SHM) systems. These systems continuously evaluate the state of the structure, allowing the real-time identification of BVID, and the estimation of the remaining service life according to the type of performance of the aircraft. If the damage severity is below a previously established value, then the component is kept operating. Therefore, an

effective SHM system minimizes the ground time for inspections, increases the availability, and allows a reduction of the total maintenance cost by more than 30% for an aircraft fleet [3].

Besides SHM, NDT techniques are applied to research, manufacturing and quality control, and aircraft maintenance [4]. Although there are some differences in the requirements for each kind of application, generally they are all used to evaluate material properties, defects and anomalies, and to assess the capability of a structure to perform a certain function/task.

One of the first NDT technologies to be successfully applied around 1955/56 was the ultrasonic C-scan [5]. Over the years, several improvements were implemented and today it is “the primary inspection method for composite materials” [6]. In 1992, Galea and Saunders [7] developed an in-situ C-scanning system to monitor damage growth in composite specimens during fatigue tests, without removing the coupons from the loading machine. In a manufacturing perspective, Kas and Kaynak [8] used C-scan to evaluate microvoids inside composite plates produced by resin transfer moulding (RTM). More recently, Hasiotis et al. [9] detected artificial defects in laminates with ultrasonic C-scan.

According to Hung [10], digital Shearography has been receiving considerable industrial acceptance as a laser-based NDT method for full-field inspection of composite structures. Therefore, some researchers have explored its advantages and limitations, using several different approaches. Amaro et al. [11] compared the performance of electronic speckle pattern interferometry (ESPI), ultrasonic C-scan and Shearography in the

2

detection of BVID in composite laminates. In a similar perspective, Ruzek et al. [12] assessed impact damage in carbon sandwich panels from an all-composite aircraft wing using Shearography and ultrasonic C-scan. He reported that, for the tested application, Shearography was the most suitable method. Steinchen [13] focused on the advantages of using a small and mobile measuring device in conjunction with image processing software, turning Shearography into an NDT method with online, full-field and non-contact capabilities that can be easily employed in field/factory environments.

The NDT methods for an SHM system should be capable of reliably detecting the damage-induced changes in local and global properties, which are encoded in the dynamic response of the structure. Among them the Lamb wave method has been reported as “one of the most encouraging tools for quantitative identification of damage in composite structures” [14]. In a quality and process control perspective, Habeger et al. [15] studied the propagation of ultrasonic plate waves in order to evaluate their capability in measuring paper strength. Also in a production monitoring orientation, Miesen et al. [16] demonstrated it is possible to detect flaws in one sheet of unidirectional CFRP prepreg by capturing Lamb waves with conventional piezoelectric sensors. To better understand the physical phenomena, Percival and Birt [17] developed and validated a one-dimensional finite element model in order to solve the equations for the propagation of Lamb waves in anisotropic laminates. Using the fundamental symmetric Lamb mode, Birt [18] successfully evaluated delamination and impact damage in carbon-fibre laminates. Later, Grondel et al. [19] developed a SHM system using Lamb waves and acoustic emissions to detect impact and debonding damages in a composite wingbox. The extraction of signal characteristics can be hindered by the complexity of Lamb wave propagation phenomena. Therefore, to make the identification process easier, Kessler et al. [20], Grondel et al. [21], Su and Ye [22], Diamanti et al. [23], and Giurgiutiu and Santoni-Bottai [24] have designed several different systems of multi-element piezoceramic (lead zirconate titanate piezoelectric ceramic material, PZT) wafer transducers for optimal and selective generation of damage-sensitive Lamb modes, enabling more accurate damage detection in composite plates.

The main goal of this study was to assess the

suitability of the Lamb wave method, in particular of the fundamental Lamb modes, to detect three different levels of multiple BVID on carbon-epoxy composite plates, and, if possible, to improve its diagnosis capabilities. Digital Shearography with thermal loading and ultrasonic C-scan were used to substantiate the results from the Lamb wave tests. The comparison between these two additional NDT methods is expected to yield important conclusions about their sensitivity to BVID, and contribute to an improvement of the quality control capability, which is also a means to increase structure reliability.

2. NDT methods 2.1 Lamb wave method 2.1.1 Lamb wave response optimization

When a PZT wafer is bonded to a structure as an actuator, the coupling between the piezoelectric material and the specimen enables the transmission of vibrations. The simultaneous strains along the three wafer dimensions, generated through the converse piezoelectric effect [25], induce shear stresses which excite multiple Lamb wave modes. This unavoidable fact poses a problem, because it is not possible to produce a pure Lamb wave mode, and therefore the interpretation of the waveform is more difficult.

Nevertheless, the selective generation of Lamb wave modes can be improved by using a multi-element approach. By mounting a pair of rectangular PZT transducers side by side, as in Fig. 1a, it is possible to enhance the amplitude of a specific mode if the inter-element distance, ie, is set as a multiple of the wavelength [21]. Therefore, this approach implies previous knowledge of the phase velocity in order to calculate the wavelength.

a) b) Fig.1 – Improvement of the Lamb wave mode selection, using a)

rectangular PZT wafer transducers, and b) circular PZT wafer transducers

An alternative approach is to mount a pair of circular PZT transducers on both specimen surfaces, as depicted in Fig. 1b. In this case, if the pair is excited in phase, symmetric modes are preferably generated. On the contrary, if the pair is excited in anti-phase, the predominantly generated Lamb modes are anti-symmetric [22, 25]. This approach depends only on the excitation frequency.

Besides tuning the Lamb wave mode selection, it is also crucial to minimize the dispersion phenomenon. For that, the actuation signal parameters, such as frequency, amplitude, number of cycles and pulse shape, have to be optimized. The choice of the frequency has to take into account three aspects [21]:

1) The number of Lamb modes should be as small as

possible 2) The Lamb modes should be as non-dispersive as

possible 3) The wavelength should be equal to or smaller than

the size of the damage to be detected These requirements can be addressed by looking at

the dispersion curves. The theoretical curves in Figs. 2 and 3 were calculated by the Vallen Dispersion program, version R2001.0806, from Vallen-Systeme GmbH, by inserting the thickness of the specimens and the velocities of longitudinal and transverse waves in Table 2. The

3

theoretical curves in Fig. 4 were determined using the formalism presented by Kessler et al. [20]. It is relevant to mention that the curves in Fig. 3 agree with those in Fig. 4 for the frequency range 0 – 650 kHz.

To satisfy the first condition, only the zero-order modes, S0 and A0, should be generated. The second condition implies that, at the driving frequency, the slope of the group velocity dispersion curves should be nearly zero, so that the group velocity is less frequency dependent, and the dispersive effect of the propagation distance upon the waveform can be avoided. The third aspect has a relative importance. In a real application, the size of the damage is one of the things to be evaluated, and therefore it is unknown. However, when investigating the Lamb wave method itself, the damage size is a controlled parameter, and thus it should be taken into account in order to achieve the best assessment of the NDT technique. Either way, it is true that the smaller the wavelength, the higher the sensitivity of the Lamb mode.

Fig. 2 - Phase velocity dispersion curves for an isotropic material with cl

= 7199.55 m/s, ct = 3367.60 m/s, and a thickness of 2.24 mm

Fig. 3 - Group velocity dispersion curves for an isotropic material with cl

= 7199.55 m/s, ct = 3367.60 m/s, and a thickness of 2.24 mm

Fig. 4 - Comparison between experimental and theoretical dispersion

curves

After pondering all the selection criteria, the chosen excitation frequency for damage detection was 500 kHz. At that frequency, the group velocity curve in Fig. 3 shows that only the S0 and A0 Lamb modes exist, displaying nearly non-dispersive behaviour (plateau regions). According to Fig. 2, for the A0 mode, the phase velocity at 500 kHz is 2555 m/s, yielding a wavelength of 5.11 mm. At the same frequency, the S0 mode has a phase velocity around 5827 m/s and a wavelength of 11.65 mm.

The signal amplitude is directly related to the magnitude of the Lamb wave strain. Therefore, higher amplitude induces a higher signal-to-noise ratio, yielding a clearer signal. The voltage amplification should be limited, because, at a certain point, the signal drift begins to deteriorate the resolution of the acquisition system [20].

The number of cycles is one of the most important parameters, because it has direct influence on the frequency content of the signal. Fig. 5 presents the time-domain representations of a one-cycle sinusoidal burst at 500 kHz and a five-cycle sinusoidal burst at 500 kHz, as well as their frequency-domain representations calculated through the Fast Fourier Transform (FFT). The comparison between Figs 5a and 5b clearly demonstrates that the larger the number of cycles, the narrower the bandwidth, and therefore the less dispersive is the Lamb wave propagation. The bandwidth can be further reduced if a Hanning function is used to window the original sinusoid, producing an N-cycle amplitude-modulated tone-burst at central frequency f. In Fig. 6b, where a five-cycle windowed sinusoidal tone-burst at 500 kHz is represented in both time and frequency domains, it is clear that, although the windowing process reduces the energy content of the signal, it eliminates the side frequencies almost completely. So, it is advantageous to use a higher number of cycles, because it reduces dispersion and increases the energy dedicated to the desired frequency, implying a clearer and stronger Lamb wave response [20]. This way, the group velocity calculation is more accurate, increasing the sensitivity and reliability of the damage detection system.

However, in a short specimen, a burst with more cycles induces a more complicated reflection pattern,

4

which makes the Lamb wave response more difficult to interpret [14, 20]. Hence, the alternative solution was to increase the signal amplitude, while keeping only one cycle. Fig. 6a shows the time and frequency contents of the one-cycle sinusoidal burst from Fig. 5a amplified five times. It is possible to observe that, instead of changing the frequency content of the excitation pulse, this operation increases the energy dedicated to the desired frequency, enhancing the Lamb wave strain.

As for the pulse shape, it has been proven [20, 21, 22, 23] that sinusoidal waveforms are able to excite Lamb wave modes more efficiently, especially if used as windowed tone-bursts.

Fig.5 - Time and frequency domain contents for pure sinusoidal bursts

with a) 1 cycle, and b) 5 cycles

Fig.6 - Time and frequency domain contents for sinusoidal tone-bursts, with by a) 1 cycle and modified by an amplitude increase, and b) with 5

cycles modified by a Hanning function windowing process

2.1.2 Damage detection

The presence of BVID originates more discontinuities within the composite material and reduces its stiffness. So, when Lamb waves propagate through a damaged area, there are more reflections and transmission of mechanical energy is less efficient [20], when compared to the healthy material. This physical evidence of the

presence of damage is encoded as variations of signal parameters.

An increase in the number of discontinuities can be translated into a stronger signal attenuation, which is directly evaluated by measuring and comparing the amplitude of the healthy and damaged responses [23]. If a stiffness reduction is induced, the group velocity will become lower. Experimentally, the group velocity is calculated by dividing the propagation distance by the travelling time (Time-of-Flight, TOF) of the wave packet. So, the presence of damage can be translated into a lower group velocity, which can be evaluated by detecting the change in TOF [16]. 2.2 Digital Shearography

Lasers emit highly coherent (linearly polarized) light. When the light is randomly scattered from the object surface and passes through a modified Michelson interferometer, a second wave front is generated, creating two laterally sheared images [26].

Due to the image shearing process, rays from neighbouring object points are superimposed. So, because each one of those rays has different phase, spatial interference occurs and an interferometric speckle pattern is created [26]. This pattern is called interferogram. The intensity of the interferogram depends on the phase difference.

When loading is applied, the object is deformed, and the surface points occupy new positions. In the deformed state, the phase distribution of the randomly scattered light is different from the phase distribution in the undeformed state. Therefore, the new interferogram has a different intensity, whose value depends on the relative phase change, ∆ [26].

The digital subtraction of the two interferograms yields an image called shearogram, whose intensity also depends on the phase change. The spatial variation of ∆ defines a pattern of bright and dark areas called the fringe pattern. Hence, the fringe lines are the loci of points where the intensity is at a minimum or maximum [26]. If the value ∆ = 2π is assigned to white pixels, and ∆ = 0 to black pixels, then there are 256 phase levels for an 8-bit resolution system.

The phase change carries information about the deformation field. An out-of-pane Shearography instrument has the illumination direction normal to the object surface. So, the presence of fringes depicted in a shearogram can be interpreted as out-of-plane strain concentration areas [26]. Any discontinuity within the material changes the strain concentration field around it. So, the evaluation of the ∆ distribution in a shearogram allows the quantification of the changes in the out-of-plane strain concentration, enabling the damage detection. The damage size that can be identified depends on the sensitivity, which can be enhanced by increasing the shear amount [26].

5

2.3 Ultrasonic C-scan

Ultrasonic C-scan is an NDT method that uses ultrasounds to generate an “overall view of the planar extent of damage” [6]. The through-thickness C-scanning technique measures the signal attenuation after propagation through the entire thickness, “at each particular location on the specimen.” [8] The energy loss is a function of the amount of reflection and scattering phenomena that occur when the ultrasonic waves encounter discontinuities. Thus, when the signal attenuation is higher than the value for the surrounding material, it means those non-uniformities are damage sites. This information is then displayed as a coloured dB map of the inspected surface. 3. Experimental procedure 3.1 Manufacturing of the composite specimens

The material selected for this study was the M30SC/120DT carbon/epoxy uni-directional (UD) prepreg, supplied by Delta-Tech S. p. a. The prepreg properties are presented in Table 1. The 7-ply arrangement [+45/-45/0/90/0/-45/+45], based on a McDonnel Douglas design [27], was adopted as a representative stacking sequence for an aircraft structure. The lay-up sequence was composed of two 7-ply blocks, [+45/-45/0/90/0/-45/+45]2, in order to produce a laminated plate with 410 x 410 x 2.24 mm3. The material was cured in the autoclave, according to the material supplier recommendations. The cured composite plate was then cut into four 400 x 200 x 2.24 mm3 specimens.

The mechanical properties of the composite specimens listed in Table 2 were calculated according to the Classical Laminated Plate Theory (CLPT) [28]. Although the ratio between Ex and Ey revealed a level of anisotropy of 29%, the laminate was considered quasi-isotropic and the average of Ex and Ey was taken as the Young’s modulus of the material. With that value, the longitudinal and transverse wave velocities, cl and ct respectively (Table 2), were computed [20].

Table 1 - Properties of the M30SC/DT 120 UD prepreg [29]

E11 (GPa)

E22 (GPa)

G12 (GPa)

ν12 (-)

Fibre fraction

(%)

Thickness (mm)

Density (kg/m3)

155 7.8 5.5 0.27 66 0.16 1760

Table 2 - Mechanical and acoustic properties of the specimens

Ex (GPa)

Ey (GPa)

νxy (-)

νyx (-)

cl (m/s)

ct (m/s)

63.5 45.1 0.29 0.42 7199.55 3367.60

3.2 Quality control and ultrasonic C-scan tests

The final quality of the specimens was assessed by performing an ultrasonic C-scan. For that, a Midas NDT Systems C-scan was used in the through-thickness configuration, using two water jets to transmit the 10 MHz signals. This measurement allowed the evaluation of the health of the specimens, which was crucial for the choice of the reference undamaged plate. After analyzing the

results, it was observed that the dB standard deviation for the four samples was lower than for the C-scan standard panel. Furthermore, specimen 1 had the lowest dB standard deviation among all the four plates, meaning it had the lowest signal attenuation, and therefore less internal defects (better quality). For this reason, specimen 1 was chosen as the reference sample for the undamaged condition of the material.

After inducing the multiple low-energy impacts (section 3.3), the damaged specimens were tested with the same Midas NDT Systems C-scan in the through-thickness configuration, using the same 10 MHz ultrasonic transducer. The dB maps of the signal attenuation were evaluated by the Automated Laminate Inspection System (ALIS) software package.

3.3 Application of multiple low-velocity impact damage

It was decided to apply ten impact points randomly distributed inside a 6 cm radius circle. So, ten known random positions were generated by MATLAB®, forming the pattern depicted in Fig. 7. The multiple BVID was induced to specimens 2, 3 and 4, always using the same impact pattern. The impactor had a mass equal to 1.2 kg and its hemispherical head had a diameter of 12.7 mm.

Fig. 7– Impact pattern

In order to test different damage severities, impact

energies of 3, 5 and 10 J were applied to specimens 2, 3 and 4, respectively. The energy level of each impact was set by adjusting the height from which the impactor would be released, according to the definition of gravitational potential energy. Thus, the 3, 5 and 10 J impact damages were applied by releasing the impactor from 25.5, 42.5 and 85 cm, respectively. It was not possible to apply an impact energy lower than 3 J, because the impact tower could not release the impactor from heights lower than 25 cm. 3.4 Digital Shearography tests

Each specimen was placed 800 mm away from an Isi-Sys 5-MPixel-CCD camera, with an inclination approximately equal to zero degrees with respect to the axis of a Nikon 50 mm lens. Four lasers of 100 mW, with a wavelength of 658 nm, were placed on each side of the

6

camera (eight lasers in total), illuminating the back surface of the specimen, as depicted in Fig. 8.

Fig. 8 - Digital Shearography setup

The thermal load was applied by a 500 W halogen

lamp, and three different loading times, 5, 10 and 15 seconds, were tested. A shear of 2.5 mm was applied in the x and y directions by adjusting the tilting of the mirrors of the modified Michelson interferometer. The Shearwin software was used to filter and demodulate the captured images, allowing the evaluation of the shearograms. 3.5 Lamb wave tests

It was decided to use two circular PZT wafers to predominantly generate the S0 mode (S0 configuration), and two rectangular PZT wafers for the selective excitation of the A0 mode (A0 configuration). Having obtained the estimate for the wavelength of the A0 mode at 500 kHz, the inter-element distance, ie, was set equal to 2 x 5.11 = 10.22 mm.

The Physik Instrumente PZT wafers were bonded to the surface of the specimens, using silver-loaded-conductive epoxy. The actuators in the S0 and A0 configurations were positioned in opposite ends of the composite plate, according to the distances defined in Fig. 9 and listed in Table 3.

An Aglient 33220A arbitrary waveform generator was used to produce a 1 cycle-sinusoidal tone-burst with 400 mVpp. The electrical signal was then amplified by an E/N 2100L RF 50dB power amplifier and transformed into mechanical waves by the PZT wafer actuators, either in the S0 or in the A0 configuration. The response of the specimen was captured by two Physical Acoustics Pico HF-1.2 micro-miniature PZT sensors in the pitch-catch configuration, glued by Olympus Sonotech shear gel and strong adhesive tape. To complete the setup depicted in Fig. 10, the signals were amplified by two 40 dB in-house-built pre-amplifiers, and acquired by a digital oscilloscope PicoScope 4424, from Pico Technology, which was operated by the PicoScope 6 software. The excitation signal was transmitted from the power amplifier to channel A through a Yokogawa 700997 cable that reduces the amplitude by ten times in order to protect the digital oscilloscope from overcharges.

The actuators and the sensors were chosen such that their resonance frequencies were close to the excitation frequency, in order to avoid non-linear phenomena in the signals (see Tables 4 and 5).

Fig. 9 - Distances between actuator and sensor, and between two sensors

Table 3 - Values of the distances in Fig. 9, for each specimen Specimen

Distance Reference 3 J 5 J 10 J D(A,S)A (cm) 4.25 4.2 4.2 4.3 D(A,S)S (cm) 4.2 4.2 4.2 4.3 D(S,S) (cm) 14 14 14 14

Fig. 10 - Setup for Lamb wave measurements

Table 4 – Properties of the Physik Instrument PZT wafer actuators. [30]

Properties Rectangular PZT Circular PZT Material PIC 151 PIC 255

Dimensions (mm)

Length (L) = 20 Width (W) = 5 Diameter (D) = 10

Thickness (TH) = 1 Thickness (TH) = 1 Resonance Frequencies (kHz)

f(L) = 100 f(W) = 400 f(D) = 200

f(TH) = 2000 f(TH) = 2000

Table 5 – Properties of Physical Acoustics Pico HF-1.2 PZT sensors [31]

Dimensions (mm) Diameter (D) = 5

Thickness (TH) = 4 Weight (g) 0.1 (7 with cable and connector) Operating frequency range (kHz)

500 – 1850

Resonance frequency (kHz)

550

The onset of the excitation signal in channel A was

defined as the repetitive trigger for capturing the two response signals in channels B and C. The sensor closest to the active actuation unit was connected to channel B, which was then called channel 1. The sensor further away from the actuators was connected to channel C, which was

7

then denoted channel 2. The active actuation unit was changed from S0 configuration to A0 configuration when the desired dominant mode was changed from S0 to A0.

For the undamaged reference specimen, Lamb wave measurements were performed with both actuator configurations, at frequencies ranging from 100 kHz to 2000 kHz, with increments of 100 kHz. For the damaged specimens, Lamb wave measurements were performed with both actuator configurations, only at a frequency of 500 kHz. In all the tests, only 1 cycle was used in order to make the Lamb wave response easier to interpret.

For each measurement, at each frequency, the PicoScope 4424 acquired amplitude data of 32 sets of waveforms from the three channels. The TOF between the actuator and each sensor was extracted through the onset times of the acquired signals picked according to the Akaike Information Criterion (AIC) [32], and then used to divide the corresponding known propagation distance, yielding the group velocity of the desired Lamb mode. During the measurements, the noise in the acquired signals was reduced by using the PicoScope 6 resolution enhancement tool, which uses a moving-average filter [33]. An effective resolution of 14 bits was chosen. The difference between that number and the baseline 12-bit vertical resolution of the PicoScope 4424 was automatically used to set the size of the moving-average filter equal to 16 bits. 3.6 Optical microscope observations

To finalize the tests, the surface of the damaged specimens was observed through an Axiovert 40 MAT optical microscope, with Carl Zeiss lenses (2.5 and 10 times magnifications), and a 100 W illuminator. 4. Results and discussion 4.1 Lamb wave method

According to theory [17, 22, 25], three wave groups are expected to occur at 500 kHz, corresponding to the zero-order symmetric Lamb mode (S0), the zero-order shear-horizontal mode (SH0), and the zero-order anti-symmetric Lamb mode (A0), respectively.

After carefully analyzing the waveforms, it was concluded that the first wave group is the S0 Lamb mode, and the third corresponds to the A0 mode. Therefore, the S0 mode was investigated in both channels for the S0 configuration, and the A0 mode was investigated only in channel 2 for the A0 configuration.

The TOF between the actuation unit and each PZT sensor was set equal to the arrival time for each channel, and then used to divide the corresponding known propagation distance, yielding the group velocity of the desired Lamb mode. For the sake of statistical relevance, it was decided to analyse 7 sets of waveforms for each specimen, for each actuator configuration, and for each frequency. The statistical analysis was performed using the mean value, the standard deviation, and the coefficient of variation, CV (the ratio of the standard deviation to the mean value) [34]. So, the group velocity at each tested

frequency was established as the mean value of the 7 extracted values. This procedure was applied to determine the experimental dispersion curves for the healthy specimen, in Fig. 4. The S0 velocities show good agreement with the theoretical results, while the A0 velocities show a larger deviation from the predicted curve. This is probably related to the nature of each Lamb wave mode. The S0 particle motion is quasi-axial, while A0 has a quasi-flexural displacement field. Moreover, contrary to isotropic materials, group velocity is direction-dependent in composite plates. Thus, the S0 mode energy can be more easily propagated than the energy of the A0 mode, because there are fibres in the plane of propagation. The absence of fibres along the flexural displacement field yields an A0 group velocity lower than in isotropic material. The CV for the velocity measurements varied between 0.07% and 0.38% for the S0 configuration, and between 0.03% and 0.85% for the A0 configuration.

The damage detection was performed by evaluating the changes in group velocity and amplitude. The quantitative evaluation of the changes was done according to the Damage Index approach [3], by defining the lag coefficient and the attenuation coefficient in equations (1) and (2). The lag is based on the ratio between the damaged and undamaged group velocities, vd and vu, respectively. The attenuation uses the ratio between the damaged and undamaged signal amplitudes, Ad and Au, respectively. This way, it was possible to relate the reduction in group velocity and amplitude with damage severity.

��� = 1 −��

� (1)

��� ���� = 1 −��

� (2)

The group velocity and amplitude were determined

using the same statistical approach. For the A0 configuration the lag and attenuation coefficients from channel 2 are plotted in Fig. 11a. For the S0 configuration, the coefficients from channels 1 and 2 are plotted in Figs. 11b and 11c, respectively.

For the A0 configuration, the 3 and 5 J damages produced very small group velocity variations (3.1% and 0.3%, respectively), but for 10 J a lag coefficient of about 30.9% was obtained. With the S0 configuration, the 3 J level was practically not detected in both channels, because the lag coefficient remained approximately zero. However, for 5 and 10 J, both channels detected clear reductions of S0 group velocity. In channel 1 the lag coefficient was around 23.7% for both energies, while in channel 2 it was 9.9% and 11.7%.

Contrary to the group-velocity-based-detection, the evaluation of the attenuation coefficient allowed the detection of the three damage levels with the A0 configuration, yielding 28.9%, 31.1% and 60.6% for 3, 5 and 10 J, respectively. For the S0 configuration, the 3 J impacts remained practically undetected, with 9.5% and 0.5% for channel 1 and 2, respectively. Only the 5 and 10 J energies were detected, with 54.8% and 87.95% for channel 1 and 2, and 46.5% and 95.6% for channels 1 and 2, respectively. So, although all the damage energies are

8

detected with A0 configuration, the S0 configuration presents higher attenuation coefficients for 5 and 10 J, and a clearer difference between levels, which can be observed in the Lamb wave response example of Fig. 12. It is important to mention that for this quantitative evaluation, only channel 2 was used for the A0 configuration, because it was not possible to unambiguously identify the A0 mode in channel 1.

a) b)

c)

Fig. 11 - Lag and attenuation coefficients for a) A0 configuration, channel 2, b) S0 configuration, channel 1, and c) S0 configuration, channel 2

The best results for the S0 may indicate that BVID

occurred deeper in the plate thickness, since the S0 mode is expected to be more sensitive to in-depth damage, while the A0 is expected to be more sensitive to surface damage [14, 25]. This would be a justification for the results obtained with the A0 configuration.

Fig. 12 - Signals from channel 1, for the S0 configuration at 500 kHz, for

impact energies of a) 3 J, b) 5 J, and c) 10 J

An interesting fact to discuss is the difference between single and multiple BVID in composites, and the effect of those differences in the detection results.

According to Diamanti et al. [23], a single 8 J impact on a plate with 2.3 mm of thickness, produced a damage area of approximately 225 mm2, and an amplitude ratio of about 18%. Grondel et al. [21] were able to detect an almost totally attenuated Lamb wave signal, after inducing a single 24 J impact on a plate with 4.78 mm of thickness. In the present study, the multiple 5 J impacts yielded an attenuation of 54.8% for channel 1 in the S0 configuration. Therefore, it seems valid to say that multiple BVID has a more severe effect on the structural integrity of the composite plate than one single impact at slightly higher energy. 4.2 Digital Shearography

The 3 J impacts were not detected for any shear or loading time. So, it means the 3 J damage did not alter the out-of-plane strain distribution. The detection of the 5 and 10 J damages proved that those impact energies were able to produce internal damage (through-thickness damage) on the specimen, since the Shearography measurements were performed on the back surface. In a more detailed analysis of the 5 and 10 J results, the damage was quantitatively evaluated by visually measuring the approximately circular area of the impact points. The areas of the identified points were averaged, yielding the mean value of the damage area for each energy level.

The presence of 5 J damage was not entirely detected. With shear in the x direction, impacts 1 and 3 were not identified by the 5 s loading, and impact 3 was not identified by the 10 and 15 s loadings. A similar thing a happened with y shear, for which points 1, 2 and 3 were not visible after a 5 s heating, and points 1 and 3 were not visible after a 10s heating. However, for a heating time of 15 s, the y shear configuration was able to detect all the impact points. Hence, it seems that, for 15 s heating and shear in the y direction (direction of the 90º fibres), shearograms are more sensitive to changes in the concentration of the out-of-plane strain induced by the applied multiple BVID. The existence of undetected damages also suggests that there are impacts, namely points 1 and 3, which were less severely applied than others.

Fig. 13 - Shearograms for the 10 J specimen, with x shear, loaded 15 s

For the 5 J damage, the average damage area for x

and y shears remained approximately equal to 0.15 cm2 and 0.20 cm2, respectively, as the heating time increased. The average values and the CV were always higher for the y shear, which may have two interpretations. Higher mean value may indicate that the y shear allows an evaluation of

9

the real damage severity. On the other hand, higher CV indicates more variability in the measuring process, and thus a less reliable evaluation.

The 10 J impacts were all detected, with both x and y shears, and for all the heating times, as depicted in Fig. 13. For points 4 and 5 there were overlapping damage areas, which may imply coalescence of internal delaminations. The average damage area increased with loading time. With x shear, the average damage area was 0.41, 0.50, and 0.66 cm2, for 5, 10, and 15 s heating, respectively. With y shear, the average damage area was 0.51, 0.74, and 0.75 cm2, for 5, 10, and 15 s heating, respectively. In this case, the trend was different from the 5 J case, because, although the average values were higher for the y shear, the CV was lower for the y shear. Thereby, the evaluated damage severity is closer to reality and the variability in the measuring process is lower, yielding a more reliable evaluation. Again, this seems to suggest that the y shear is more sensitive to multiple BVID, especially if combined with moderate loading times. 4.3 Ultrasonic C-scan

As with Shearography, the 3 J BVID was not detected and it was possible to observe a clear distinction between the 5 and 10 J impact energies. Furthermore, for each of the detected energies, there were some impact points more severe than others. In Fig 14, the 10 J impacts have a characteristic purple ring of more severely damaged material around a small blue, less damaged region. This occurrence was confirmed by the optical microscope observations of the plate surface in section 4.4.

Fig. 14 - Attenuation maps of the 10 J damaged specimen

In a quantitative analysis, the evaluation was done by

measuring the approximately circular area around each impact point, similarly to the approach used for Shearography. Impact point 3 stands out from the others, because it has the smallest damage area, confirming it is the less severe of all. That is why it was the most difficult point to identify in Shearography tests. The C-scan mean values of the measured damage areas for 5 and 10 J, 0.33 cm2 and 1.03 cm2 respectively, are larger than the corresponding values for all the Shearography tests. Additionally, the C-scan CV values for 5 and 10 J are very similar to the values obtained with Shearography. So, it is possible to conclude that the C-scan allowed the detection of a more real damage severity, with the same level of reliability as Shearography.

Nevertheless, the relatively high coefficient of variation for the C-scan and Shearography measurements

denotes the existence of errors in the application of the impact energy. The relatively high CV may also indicate that the damage-area-measuring technique is not accurate enough, since it is based on visual inspection of the acquired images.

For 10 J specimen impact points 4, 5 and 6 had overlapping damage areas. This overlapping may be caused by the proximity of the small surface cracks (see section 4.4), or by the coalescence of internal delaminations.

4.4 Optical microscope

Microscopic observations of the surface of the damaged plates were performed in order to obtain data merely regarding the surface damage. Only one impact point was observed for each damaged specimen, so no statistical study could be made. The picture for one impact damage on the surface of the 5 J specimen is shown in Fig. 15.

Fig. 15 - Microscopic observation of one impact damage on the surface of

the 5 J damaged specimen

It was very difficult to discern the presence of a 3 J impact. Only a slight circular pattern discloses the barely-visible damage. For the 5 and 10 J specimens a circular border of what seems to be crushed material surrounds an undamaged region, forming a valley with approximately the shape of the hemispherical impactor. The areas of the circular valleys show that the indentation area is slightly larger for the 10 J damage.

More importantly, the 5 and 10 J impacts have two small surface cracks along the 45 degree direction. This specific direction is not a random occurrence. It shows that is the direction of highest stress concentration during the impact, promoting the rupture of the fibres in the surface layer. The cracks are longer for the 10 J impact. The cracks have “cliffs” on both sides of the cracks, creating shadow on the outer part. These small up lifting cracks probably conceal internal delaminations underneath, because they are not confined to the surface layer. Hence, the growth of the small cracks seems to be the cause for the increase in the measured damage area from 5 to 10 J in the Shearography and in the C-scan results. 5. Conclusions

The results of this experimental study about the Lamb wave method showed that the 3 J damage did not induce significant variations on the composite plate response, remaining practically undetected, whilst the 5 and 10 J damages were clearly detected by both actuator configurations. The amplitude-based approach produced more unambiguous results, as it was possible to detect an

10

almost totally attenuated signal for the 10 J case. Taking into account the theoretical model, the piezoelectric transducers, and the signal processing tools used, the S0 configuration proved to be more sensitive to the presence of multiple BVID. When compared with previous studies [51, 53], the Lamb wave results showed that the presence of multiple BVID has a more detrimental effect on structural integrity than one single impact at a slightly higher energy.

In the digital Shearography measurements, shear in the x and y direction was combined with three different heating times, in order to study the effect of the loading time on the detection capabilities of the shearograms. None of the configurations revealed the 3 J impacts. The 5 J multiple BVID was partially detected by the x shear, and only with y shear and 15 s heating it was possible to identify all the 5 J damages. The 10 J impacts were all detected, for both configurations and all the loading times, with the combination of y shear and 15 s heating having the best results.

The ultrasonic C-scan was able to detect the 5 and 10 J damages, and confirmed the higher reliability of the results from the combination of y shear with 15 s loading. Both the Shearography and the C-scan images showed some overlapped impact points, probably revealing subsurface delaminations. This suspicion was then substantiated by microscopic observations of the plate surface. So, it was concluded that, for the tested specimens and experimental setup, Shearography can be as reliable as C-scan, as long as appropriate loading is applied. At the end, the Lamb wave results for the S0 configuration, together with the Shearography and the C-scan images formed a coherent set of NDT results regarding multiple BVID.

The three techniques seem to have the same detection threshold, somewhere between 3 J and 5 J. Nevertheless, it also appears that the detection threshold is not imposed by the measuring technology, but by the material. Therefore, a detection threshold can only be defined with respect to a property of the material. References

[1] Beukers, Adriaan (1992), “A new technology for sandwich plates and structures based on the use of in-situ foamable thermoplastic films”. 37th International SAMPE Symposium and Exhibition, Anaheim (CA), 9-12 March 1992, Proceedings (A93-15726 04-23), pp. 172-185.

[2] Schijve, Jaap (2004), “Fatigue of Structures and Materials”. Kluwer Academic Publishers, Dordrecht.

[3] Su, Zhongqing and Ye, Lin (2009), “Identification of Damage Using Lamb Waves: From Fundamentals to Applications”. Springer.

[4] Yolken, H. Thomas and Matzkanin, George A. (2009), “Nondestructive evaluation of advanced fiber reinforced polymer matrix composites – A technology assessment”. NASA Contractor Report 2009-215566, NASA Langley Research Center, Hampton, Virginia (USA).

[5] Cartz, Louis (1995), “Nondestructive Testing – Radiography, Ultrasonics, Liquid Penetrant, Magnetic Particle, Eddy Current”. ASM International.

[6] Fahr, A. and Kandeil, A. Y. (1992), “Ultrasonic C-scan inspection of composite materials”. Engineering Journal of Qatar University, vol. 5, pp. 201-222.

[7] Galea, S. C. and Saunders, D. S. (1992), “In-situ ultrasonic C-scanner”. Technical Report 4, Aeronautical Research Laboratory, Melbourne, Victoria (Australia).

[8] Kas, Y. Onur and Kaynak, Cevdet (2004), “Ultrasonic (C-scan) and microscopic evaluation of resin transfer molded epoxy composite plates”. Polymer Testing, vol. 24, pp. 114-120.

[9] Hasiotis, T.; Badogiannis, E. and Tsouvalis, N. G. (2007), “Application of ultrasonic C-scan techniques for tracing defects in laminated composite materials”. 4th International Conference on NDT, Chania, Crete (Greece), 11-14 October 2007.

[10] Hung, Y. Y. (1999), “Applications of digital Shearography for testing of composite structures”. Composites, Part B: Engineering, vol. 30, pp. 765-773.

[11] Amaro, Ana M.; Santos, J. B. and Cirne, José (2004), “Comparative study of different non-destructive testing techniques in the characterisation and quantification ofthe damage effects in carbon-epoxy laminates”. Insight, vol. 46, no. 9, pp. 559-565.

[12] Ruzek, Roman; Lohonka, Radek and Jironc, Josef (2005), “Ultrasonic C-scan and Shearography NDI techniques evaluation of impact defects identification”. NDT&E International, vol. 39, pp. 132-142.

[13] Steinchen, Wolfgang; Yang, Lianxiang; Kupfer, Gerhard and Mäckel, Peter (1998), “Non-destructive testing of aerospace composite materials using digital Shearography”. Proceedings of Institution of Mechanical Engineers, vol. 212, Part G, pp. 21-30.

[14] Su, Zhongqing; Ye, Lin and Lu, Ye (2006), “Guided Lamb waves for identification of damage in composite structures: A review”. Journal of Sound and Vibration, 295, pp.753-780.

[15] Habeger, Charles C., Jr.; Mann, Ronald W. and Baum, Gary A. (1979), “Ultrasonic plate waves in paper”. Ultrasonics, pp. 57-62.

[16] Miesen, Nick; Mizutani, Yoshihiro; Groves, Roger M.; Sinke, Jos and Benedictus, Rinze (2011), “Lamb wave detection in prepreg composite materials with fibre Bragg grating sensors”. Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2011, 7th March 2011, San Diego, California (USA), Proceedings SPIE 7981, 79812J.

[17] Percival, W. J. and Birt, E. Anne. (1997), “A study of Lamb wave propagation in carbon-fibre composites”. Insight, vol. 39, no. 10, pp. 728-735.

[18] Birt, E. Anne. (1998), “Damage detection in carbon-fibre composites using ultrasonic Lamb waves”. Insight, vol. 40, no. 5, pp. 335-339.

[19] Grondel, Sébastien; Assaad, Jamal; Delebarre, Christophe and Moulin, Emmanuel. (2004), “Health monitoring of a composite wingbox structure”. Ultrasonics, vol. 42, pp. 819-824.

[20] Kessler, Seth S.; Spearing, S. Mark and Soutis, Constantinos (2002), “Damage detection in composite materials using lamb wave methods”. Smart Materials and Structures, vol. 11, pp.269-278.

[21] Grondel, Sébastien; Paget, Christophe; Delebarre, Christophe and Assaad, Jamal (2002), “Design of optimal configuration for generating A0 Lamb mode in a composite plate using piezoceramic transducers”. Journal of the Acoustical Society of America, vol. 112 (1), pp. 84-90.

[22] Su, Zhongqing and Ye, Lin (2004), “Selective generation of Lamb wave modes and their propagation characteristics in defective composite laminates”. Proceedings of the Institution of mechanical Engineers, vol. 218, Part L: Journal of Materials design and Applications, pp. 95-110.

[23] Diamanti, K.; Soutis, C. and Hodgkinson, J. M. (2006), “Piezoelectric transducer arrangement for the inspection of large composite structures”. Composites, Part A: Applied Science and Manufacturing, vol. 38, pp. 1121-1130.

[24] Giurgiutiu, Victor and Santoni-Bottai, Giola (2011), “Structural health monitoring of composite structures with piezoelectric-wafer active sensors”. AIAA Journal, vol.49, no. 3, pp. 565-581.

[25] Giurgiutiu, Victor (2008), “Structural Health Monitoring with Piezoelectric Wafer Active Sensors”. Academic Press-Elsevier, San Diego, California (USA).

[26] Steinchen, Wolfgang and Yang, Lianxiang (2003), “Digital Shearography – Theory and Application of Digital Speckle Pattern Shearing Interferometry”. SPIE, Bellingham, Washington (USA).

[27] Masters, John E. (1997), “Translaminar fracture toughness of a composite wing skin made of stitched warp-knit fabric”. NASA Contractor Report 201728, NASA Langley Research Center, Hampton, Virginia (USA).

[28] Reddy, Junuthula N. (2004), “Mechanics of laminated Composite Plates and Shells – Theory and Analysis”. CRC Press, Boca Raton.

[29] Rodi, Riccardo; Campoli, Gianni; Alderliesten, René C. and Benedictus, Rinze (2009), “Characterization of the crack tip behaviour in fibre metal laminates by means of digital image correlation”. 50th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, vol. 2009-2586.

[30] “Piezo Material Data – Specific parameters of the Standard materials”. Physik Instrumente, 2011.

[31] “PICO HF-1.2 Sensor: 600 kHz Frequency, Micro-Miniature Acoustic Emission Sensor”. MISTRAS Group, Inc, 2010.

[32] Kurz, Jochen H.; Grosse, Christian U. and Reinhardt, Hans-Wolf (2004), “Strategies for reliable automatic onset time picking of acoustic emissions and of ultrasound signals in concrete”. Ultrasonics, vol. 43, pp. 538-546.

[33] Smith, Steven W. (2003), “Digital Signal Processing – A Practical Guide for Engineers and Scientists”. Newnes-Elsevier Science.

[34] Ott, R. Lyman and Longnecker, Michael (2008), “An Introduction to Statistical Methods and Data Analysis” (6th ed.). Brooks/Cole, Cengage Learning.