Damage tolerance in glass reinforced polymer laminates

Composites Science and Technology 95 (2014) 21–28

Contents lists available at ScienceDirect

Composites Science and Technology journal homepage: www.elsevier.com/locate/compscitech

Damage tolerance in glass reinforced polymer laminates Kenny W. Campbell a, Peter H. Mott b,⇑ a b

Science Applications International Corporation, La Plata, MD 20646, United States Chemistry Division, U.S. Naval Research Laboratory, Washington, DC 20375, United States

a r t i c l e

i n f o

Article history: Received 11 October 2013 Received in revised form 6 January 2014 Accepted 5 February 2014 Available online 11 February 2014 Keywords: C. Damage mechanics B. Mechanical properties A. Glass fibers D. Acoustic emission

a b s t r a c t Ten glass-reinforced, rubber-toughened polymer laminate panels were tested in 3-point bending in a series of loading–unloading cycles, with increasing deflection. Damage was quantified by the stiffness decrease, hysteresis and residual strain. The threshold for unacceptable damage occurred when the strain reached ca. 0.6%. Acoustic emission (AE) was monitored by four sensors on the compressive side of the samples; the correspondence between the damage threshold and different AE measures was explored, with hit strength (i.e., the measured area under the rectified signal envelope, or MARSE) providing the clearest correlation. Separating events with AE hits that were recorded by all four sensors (‘‘associated’’) from those recorded by three or fewer sensors (‘‘unassociated’’), distinguished matrix from fiber damage. Viscoelastic effects were identified by separating hits that occurred during loading from those that occurred during hold and unloading. Published by Elsevier Ltd.

1. Introduction In many applications with unpredictable loading, the superior properties of laminated, glass-reinforced composites are degraded by minor over-straining. Damage affects these composites differently from that of conventional structural materials, such as ductile metals or polymeric glasses. For conventional materials, an uncommon, small over-straining usually causes plastic deformation, with negligible changes to overall mechanical properties; on the other hand for composites, delamination and fiber failure lead to significant reductions in stiffness and strength. Due in part to sensitivity to damage, the design allowable for composites is low – typically 0.3% strain, 1=4 that of the fiber failure strain – effacing much of the weight and cost saving [1]. When deformation must exceed these limits, a better understanding of the scope of damage, and its tolerance, is needed to take full advantage of these materials. Acoustic emission (AE) holds promise as a method to evaluate the health of such structures. For laminated composites there are different types of damage, such as matrix cracking, fiber/matrix disbonding, delamination, and fiber breakage, occurring over different but overlapping ranges of strain. Amplitude sorting [2] has been used to discriminate these types: in tensile loading, this has been corroborated by scanning electron microscopy [3–6] and ultrasonic backscatter [7]; likewise, the correlations have been ⇑ Corresponding author. Tel.: +1 202 767 1720; fax: +1 202 767 0594. E-mail address: [email protected] (P.H. Mott). http://dx.doi.org/10.1016/j.compscitech.2014.02.004 0266-3538/Published by Elsevier Ltd.

demonstrated in three-point bending [8]. Faster computers and better software have enabled transient or modal analysis, where different waveforms have been linked to damage types [9–12]. Clustering analysis has been shown to discern damage types from AE signal characteristics [10–14]. Damage has also been related to the AE signal rise angle [15,16]. These studies have demonstrated and improved the diagnostic effectiveness of AE, and through these connections, with consideration to the attenuation [17,18], it is possible to determine the type and extent of damage in a material as it is strained. Building on these advancements, the purpose of this work is to show how AE may be used to avoid intolerable damage. The tested laminate is commonly used for large marine fabrications (e.g., mast fairings, bow domes, propulsor components, turtleback superstructures, etc. [19]) and is representative of many fiberglass composites [20,21]. During installation it can be necessary to flex an article to higher strains than what normally occurs in service, to where it is possible to generate flaws that weaken the structure. To investigate the potential for installation damage, specimens were subjected to bending cycles with increasing strain, the damage quantified by the stiffness decrease, hysteresis, and residual strain, and correlated to AE data. We consider our findings to be generally applicable to fiber reinforced polymer matrix composites. 2. Experimental The tested composite was fabricated from 12 plies of Cytec Industries Cycom 5920/1583, a rubber toughened epoxy prepreg

22

K.W. Campbell, P.H. Mott / Composites Science and Technology 95 (2014) 21–28

reinforced with E-glass fibers, which was developed for immersed saltwater use. The layers contain fibers in the 0° and 90° directions, with all plies laid up with warp directions parallel to the long axis of the sample (an all 0° warp lay-up), and autoclave cured using a proprietary heat and pressure sequence. The cured sheet was cut into panels, with dimensions 470 mm long  90 mm wide  5 mm thick. Ten panels were subjected to 3-point bending as shown in Fig. 1a, using an Instron 5500R test machine. The flexure sequence, shown in Fig. 1b, consisted of slow loading (1.0 mm/min, strain rate 1.2  105 s1) followed by quick unloading (–10.0 mm/min, strain rate 12  105 s1), and a 10 s hold time between the changes. The maximum deflection grew in each cycle: for the first cycle it was 2 mm, and increased by 1 mm in each succeeding cycle, for ten cycles (11 mm total deflection). After this the samples were flexed to failure. The visible damage occurred near the loading pin, which consisted of fiber failure (10 mm band), surrounded by delaminations (35 mm). The stress at the loading pin was found by

r¼

3Pl 2wt 2

The acoustic emission was detected by four Physical Acoustics R15I-AST piezoelectric transducers, with integral preamplifiers, connected to a Physical Acoustics 4-channel lDiSP system. The transducers had resonant frequencies of 125 kHz (plane waves) and 153 kHz (surface waves), and were secured with vinyl tape using Sonotech Ul tragel II as a coupling medium. A sketch of the hit voltage V(t) is shown in Fig. 2, identifying the threshold VT, duration tD, and hit strength found from the measured area under the rectified signal envelope (MARSE). The amplitude is (in dBAE)

A ¼ 20 log

ð1Þ M¼

6dt

Z

t1

jV P ðtÞjdt:

ð4Þ

t0

In strict terms M is not energy, but nevertheless it is proportional to the hit energy as it accounts for both amplitude and duration. Amplitude attenuation was determined from pencil lead breaks carried out at 50 mm intervals along the sample. The result, 0.033 dBAE/mm, indicates that from an event under the center load pin, where almost all damage occurs, the amplitude difference between the outermost and innermost detectors is ca. 3.1 dBAE. Given the data scatter, this difference was acceptable. Using linear beam theory, it can be shown that this loading geometry limits the transverse shear stress to 1.2% of the tensile

ð2Þ

2

l

ð3Þ

where VM is the maximum hit voltage, VR is a 1.0 lV reference at the preamplifier input, and G is the preamplifier gain, which was fixed by transducer circuitry to 40 dBAE. The amplitude threshold was set to 45 dBAE (VT = 17.8 mV), which discards the noise from the loading fixture, as determined by experiment. The hit strength (i.e., MARSE) is computed by integrating the rectified peak signal voltages |VP (t)| over the time the amplitude exceeds the threshold, i.e., the hit duration:

where P is the load, and l, w, and t are the respective loading span length, sample width and thickness. The strain data were from a Vishay CEA-06-250UW-350 gauge bonded to the tension side, which has a 6.35  4.57 mm2 sensing area. When the strain exceeded ca. 2% this gauge became unreliable, so strains beyond this value were determined from

e¼

VM G VR

where d is the center deflection. This relation was verified by comparing the deflection and strain gauge data at low strain.

(a)

89

89

LOAD

29 Sensor

5

40

Strain gauge

Ø19.1 203.2 470

(b)

to failure

DISPLACMENT (mm)

15

10

5

0 0

20

40

60

80

TIME (minutes) Fig. 1. (a) Schematic of loading. Dimensions are in mm; sample width (normal to page) was 90 mm. (b) Load pin displacement: the respective loading and unloading rates were 1.0 and 10.0 mm/min, corresponding to strain rates of 1.3  103 and 13  103 s1.

23

K.W. Campbell, P.H. Mott / Composites Science and Technology 95 (2014) 21–28

VOLTAGE

V(t)

VM

tD

V(t)

VP (t) M

VT t0

t1

TIME

Fig. 2. Schematic of acoustic emission hit. Top: voltage trace of hit V(t). Bottom: rectified trace |V(t)|, showing the threshold VT, hit amplitude VM, hit duration tD between the start t0 and finish t1 times, and the signal envelope defined by the rectified peaks |VP(t)|. The hit strength M (i.e., MARSE) is the area under this envelope (Eq. (4)).

stress under the center load pin. Thus, the number AE events arising from shear stress, at locations away from the primary damage area, are negligible. 3. Results Fig. 3 shows representative stress–strain curves, displaying the first and tenth cycles. This behavior was quite consistent between samples. For the first cycle, the loading and unloading responses are nearly identical, and the traces cannot be distinguished. By the tenth cycle the difference between loading and unloading is pronounced. The figure shows the definitions for the cycle strain ec (maximum strain in the cycle), hysteresis H (the area enclosed by the loading–unloading curves), and residual strain er (strain after unloading). Both H and er are evidence of damage, increasing from zero for the first cycle (within the measurement uncertainty), to 68.4 kJ m–3 and 0.051%, respectively, for the tenth cycle. H and er also include a small viscoelastic response, which is not damage. Given this time-dependent recovery, for a consistent comparison

200

cle

th

STRESS (MPa)

150

cy

c

10

100

ng

di

H

a Lo

g

in

d oa

the er data were taken after the 10 s hold period. Typically the samples recuperated 5–10% of the strain after unloading. Table 1 reports the stiffness and failure properties, which agreed with previous measurements [19,22]. The Young’s modulus E data were determined by fitting a third-order polynomial to the stress–strain curve, and taking the derivative of the fit at zero strain. All listed values are averages of the ten samples, and the error the standard deviation of the data. The injury from each cycle is displayed in Fig. 4. It shows averages of Young’s modulus E, and of ratios of the hysteresis H/ec and residual strain er/ec normalized by the cycle strain, as a function of cycle strain ec (error bars are standard deviation). The horizontal error bars are 5% of ec, caused by small thickness variation (±0.25 mm). Degradation becomes meaningful when E falls below the experimental scatter of the undamaged material, at ec = 0.65% (i.e., when E had decreased by 5.3%). Beyond this strain, E falls more quickly. The E data were determined by polynomial fits, as described in the previous paragraph, at the beginning of the loading cycle; conversely er/ec and H/ec were determined at the end of the cycle, so that these data are shifted to the right. The E datum at zero strain is for the unstrained (virgin) samples, and corresponds to the value listed in Table 1. In a purely viscoelastic material, the residual strain and hysteresis are proportional to the cycle strain, so the normalized quantities er/ec and H/ec are constant. Fig. 4 shows, however, that er/ec and H/ec are increasing. This is consistent with accumulating damage, and any viscoelastic relaxation, which would appear as a vertical offset, is comparatively small. The H/ec plot is smoothly increasing from the origin, but then drops when ec > 0.9%. The er/ ec plot, on the other hand, has two interesting traits: (1) negative values when ec < 0.3%, and (2) a nearly linear behavior up until ec = 0.65%, with an upward jump thereafter. We have no explanation for the former; the latter mirrors the damage trend in the Young’s modulus. Fig. 5 displays how the AE evolves: shown are the averages of the number of hits that occur in a cycle nc, cycle hit strength per hit Mc/nc, and cycle hit amplitude Ac, as a function of cycle strain ec to failure. The three plots are alike, with a sharp increase in acoustic activity occurring when ec > 0.6%. For nc there were few hits until 0.65% strain, at which point they markedly increase, and then flatten out when ec > 1.0%. Similar results were reported elsewhere [9,11,23]. The Mc/nc data averages the hit strengths in a given cycle, divided by the number of cycle hits; it initially fluctuates around 10 lV-s at low strain, and increases by 1.5 orders of magnitude when ec > 0.6%. The hit amplitude is first steady at 55 dBAE, then steps up to about 80 dBAE at 0.65% strain, and then decreases after 1.0% strain. The amplitude decrease at high strain is consistent with the mounting damage increasing the attenuation of wave energy [12], discussed further below. Fig. 6 explores how the stiffness and hit strength vary as a function of the accumulating hits n (=Rnc). As can be discerned given the scatter, the plot of E vs. n quite surprisingly shows a linear decline, even though there is a break in E vs. ec curve (Fig. 4); the slope of the fit line is 1.1 ± 0.07 GPa/log(count). The cumulative average hit strength RMc has two, nearly linear, regimes: the first up to 10 hits, and the second follows after an upward jump,

l

50

tc yc le

Un

1s

Table 1 Comparison of stress–strain results (average ± standard deviation).

0

Property

0.0

0.2

r

0.4

0.6

0.8

This work

1.0

STRAIN (%)

Fig. 3. Representative stress–strain curves displaying the first and tenth cycles. Shown are the definitions for cycle strain ec, hysteresis H, and the residual strain er.

Young’s modulus (GPa) Tensile strength (MPa) Tensile failure strain (%)

26.3 ± 1.2 408 ± 17 2.49 ± 0.06

Previous measurements Ref. [19]

Ref. [22]

17.9 ± 1.6 426 ± 25 2.75 ± 0.16

26.6 ± 0.7 407 ± 21 —

24

K.W. Campbell, P.H. Mott / Composites Science and Technology 95 (2014) 21–28

28

28

6

10

Young's Modulus

Cumulative Cycle

E

10.0 26

Hit Strength

Σ Mc (μV-s)

(GPa)

E

0.04 r c

26

(MPa)

Young's Modulus

24 4

10

7.5 24 0.02

Residual Strain

5.0

2

22

H

10

c 3

(MJ/m )

0.00

0

2.5

10

1

2

10

10

3

10

Σ nc Hysteresis

Fig. 6. Relationship of mean Young’s modulus E and cumulative hit strength RMc to the average cumulative number of hits Rnc. Error bars represent the standard deviations; horizontal error bars in the RMc data were omitted for clarity. The dashed line is a best fit to the moduli data, with a slope of 1.1 ± 0.07 GPa/ log(count).

0.0 0.0

0.2

0.4

0.6

0.8

1.0

c (%) Fig. 4. Young’s modulus E, reduced residual strains er/ec, and reduced hysteresis H/ ec, as a function of cycle strain ec. Plotted points on both the ordinate and abscissa are averages and the error bars are the standard deviation. The values for E were determined from the derivative of a third-order polynomial fit to the individual stress–strain curve at the zero strain; points represent averages of the fits, and the error bars are the standard deviation. The dashed line for E is the lower bound of the scatter for virgin samples. The dashed line for er/ec represents the linear range.

80

Amplitude

Ac (dBAE)

Mc

nc

(μV-s)

70 3

10

The sensitivity to the amplitude threshold is examined in Fig. 8. On the ordinate, values of Young’s modulus E and residual strain er are unchanged from the previous figure; adjusting the threshold shifts these data horizontally. The low strength data (within the dashed circles) show that increasing the amplitude threshold from 45 to 73 dBAE shifts the RMc abscissa values to the left; a further increase to 77 dBAE discards them altogether. At higher hit strength, all points superimpose: adjusting the threshold has no apparent effect. For the E data (upper half of Fig. 8), apparent linear behavior is preserved regardless of the threshold, with minor changes to the best-fit line. For the er data (lower half of Fig. 8), when the amplitude threshold is increased to 73 dBAE, the two distinct responses persist, with a leftward shift of the intersecting ‘corner’. Increasing the threshold further to 77 dBAE causes all data to the left of the corner to be lost.

60

Hit Strength

4. Discussion

or nc

50 1

10

Cycle Hits

0.0

0.5

1.0 c

1.5

2.0

(%)

Fig. 5. Mean hit amplitude Ac, strength per hit Mc/nc, and number nc observed within each cycle, plotted as a function of a cycle strain ec. Error bars have been omitted for clarity.

indicating intensifying hit strength. The break in this curve corresponds to discontinuities in Fig. 4 and Fig. 5, at ec = 0.65%. In an attempt to find other linear relationships between mechanical damage measures and AE, Fig. 7 displays Young’s modulus, residual strain, and hysteresis as a function of the cumulative hit strength. Again, E shows a decrease that is apparently linear without any discontinuities. On the other hand, the plots for H and er, each show two distinct responses that intersect when RMc = 62 lV-s (where ec = 0.65%), the same place as in all other figures.

The trends in the hysteresis, residual strain, and AE data demonstrate two categories of damage. In the first type the structural changes are small. In the second type the structural changes are more severe, and occur when the strain exceeds ca. 0.6%, corresponding to when the total number of hits exceed 10 and when the hit strength exceeds 100 lV-s. The change between the two types of damage was abrupt, with large increases in hit rate, amplitude, and strength. To identify the type of damage corresponding to the AE events, the events were separated into those with coincident hits recorded on all four sensors (‘associated’), and those with coincident hits recorded on three or fewer sensors (‘unassociated’). The averages of number of events nEc (not hits), event strength MEc, and event amplitude data AEc, for associated and unassociated events within a cycle, are presented as a function of ec in Fig. 9 (error bars are standard deviation). The events were identified from hits that occurred within the time for the elastic wave to travel down the sample; the amplitude and strength of the event was taken as the maximum value of the hit group. Only unassociated events occur when ec < 0.65%, but after this, the associated outnumber the unassociated events. The unassociated MEc is steady at low strains and increase only slightly when ec P 0.65%, while the associated MEc is nearly constant and 3–4 orders of magnitude higher. Likewise the associated AEc is ca. 30 dBAE greater than the unassociated AEc; and in both types the amplitudes are unchanging throughout the test.

25

K.W. Campbell, P.H. Mott / Composites Science and Technology 95 (2014) 21–28

nEc

E

2

Associated

10

(GPa) 26

1

10

Unassociated

0

10

24

3

10 400

Associated

MEc

80

(μV-s)

H

r -6 (x10 )

2

10

(kJ/m3)

Unassociated

200

40

1

10

Residual Strain Hysteresis

0

AEc

0

1

10

2

10

3

10

10

4

10

Associated

(dBAE)

0

80

5

10

Σ Mc (μV-s)

Unassociated

Fig. 7. Correlation of Young’s modulus E, hysteresis H, and residual strain er to the cumulative hit strength RMc. All plotted values are averages; on the upper plot the vertical and horizontal error bars are the standard deviations, and on the lower plot the error bars have been omitted for clarity. The dashed line in the upper plot is a best fit to the moduli data, with a slope of 0.60 ± 0.05 GPa/log(lV-s).

60

0.2

0.4

0.6

0.8

c (%) Fig. 9. Comparison between associated events (coincident hits on all four detectors) and unassociated events (coincident hits on three or fewer detectors), as a function of cycle strain ec. Top, number of events nEc; middle, event strength MEc; bottom, event amplitude AEc. Plotted points are averages over each cycle, and error bars are the standard deviations.

E (GPa) 26

24

THRESHOLD DATA

400

BEST FIT

45 dBAE 73 dBAE

r (x10-6)

77 dBAE 200

0 0

10

1

10

2

10

3

10

4

10

5

10

Σ Mc (μV-s) Fig. 8. Consequence of adjusting the amplitude threshold on the cumulative hit strength RMc. The Young’s modulus E and residual strain er are unchanged from Fig. 7. Data within the dashed circles show that increasing the amplitude threshold from 45 to 73 dBAE shifts the points to the left; a further increase to 77 dBAE causes these points to be rejected. In upper figure, lines show the least-squares fit to the data. Only one error bar is shown for clarity, representing the standard deviation.

The amplitude distributions of the associated and unassociated events are displayed in Fig. 10. As expected, most of the unassociated events were low amplitude and most of the associated events were high amplitude, but there was some overlap of the two curves. The maximum at 50 dBAE in the unassociated events reveals that the experimentally significant, low intensity events have been registered, and thus the 45 dBAE amplitude threshold is appropriate. (Setting the threshold higher will discard unassociated events; setting it lower would include events contaminated by loading fixture noise.) Also shown in Fig. 10 are the event durations tED: for low amplitude events (AE < 60 dBAE), the associated tED are longer; for medium amplitudes events (60 6 AE < 80 dBAE), the unassociated tED are longer; and for high amplitude events (AE P 80 dBAE) the two distributions follow the same trend. There were a handful of medium to high amplitude, long duration, unassociated events, contradicting the assumption that these events are necessarily weak. An explanation consistent with these data is that the unassociated events are caused by matrix cracking, while the associated events are caused by more severe damage mechanisms such as resin/fiber disbonding, fiber fracture and delamination. Many studies have correlated matrix cracking with low amplitude, low energy hits [4,5,13,24–26]. Furthermore where ec < 0.65%, the correlation between the low strength, unassociated hits and the mechanical data (Fig. 4) suggests that this type of damage enables greater deformation with only minor changes in the stiffness and residual strain. On the other hand, as the strain is increased past 0.65%, the high amplitude, high energy associated hits, dramatic

26

K.W. Campbell, P.H. Mott / Composites Science and Technology 95 (2014) 21–28

tED

nc

Duration

(μs)

Loading 2

10

10

3

Hold-unload 1

10 10

2

0

10

Unassociated Associated

nE

5

10

Number of events 10

Loading

Mc

2

(μV-s)

Hold-unload 3

10

10

1

60

1

10

80

AE (dBAE) Fig. 10. Duration tED and number nE of associated and unassociated events as a function of event ampli tude AE. Shown duration data are averages found from logarithmic values. The error bars and the shaded region are the standard deviation of the data, also determined logarithmic values.

increase in residual strain, and loss of stiffness are all consistent with fiber failure, which also has been shown in many studies [5,24,26]. This more severe damage causes more of the load to be shifted to the matrix, which increases matrix cracking as indicated by the increase in the number of unassociated hits. To further explore how damage transpires, hits that occurred during loading were separated from those that occurred during the hold and unloading parts of the cycle. Fig. 11 show these data, plotting the average number of cycle hits nc and average cycle hit strength Mc as a function of cycle strain ec. For nc, the loading and hold–unload curves are essentially constant and identical up until ec = 0.65%, at which point they both sharply increase but also diverge, with the loading hits having the greater number. Interestingly for Mc at ec < 0.65%, the hold–unload hit strength slightly decreases, but is about 6 times that of the loading hit strength. The two curves cross at ec = 0.65%, with the loading and holdunload Mc curves increasing by 4 and 2.5 orders of magnitude, respectively. The AE during hold–unload behavior reveals the importance time-dependent relaxation as a precursor to matrix cracking. When ec < 0.65%, the hit strength shows, surprisingly, that the damage that occurs with a descending load is greater than the damage with an ascending load. On the other hand when ec P 0.65%, the number and strength of AE again demonstrate that some damage occurs during the hold–unload portion of the cycle. Similar results have been reported for other composites under decreasing load [15,16]. Since at ambient temperature there should be no viscoelastic relaxation in the E-glass filaments, the hold– unload AE is evidence that resin/fiber disbonding and fiber failure transfer additional stress to the matrix. One likely source of relaxation is rubber-toughening particles in the matrix. It has been shown that the source of the toughening properties is their cavitation under negative pressure (i.e., tensile mean normal stress) [27]. Cavitated rubber particles also strongly attenuate acoustic energy [28,29]. The decrease in amplitude at high strain,

0.2

0.4

0.6

0.8

c (%) Fig. 11. Comparison between hits that occurred during loading to those that occurred during the hold and unloading portion of cycling, as a function of cycle strain ec. Top, number of hits nc; bottom, hit strength Mc. Plotted points are averages over each cycle, and error bars are the standard deviations.

>1% (see Fig. 5), is also consistent with increasing attenuation due to cavitation of the rubber particles. In metallurgical applications, structural health may be assessed with AE data taken during a proof test with increasing cyclic loading. The damage state is characterized by the felicity ratio, defined as the load where AE begins divided by the previous maximum load. When the felicity ratio is less than unity the AE has been shown to arise from crack growth, and is evidence of existing, serious damage [30–32]. By analogy the felicity ratio is defined here as

F¼

eAE ec1

ð5Þ

where eAE is the strain at which AE is observed and ec1 is the previous cycle strain. For fiber-reinforced materials that are approaching failure, F < 1 and is decreasing [33–36]. It has been proposed that when F < 1, the AE is caused by viscoelastic relaxation of the matrix after fiber failure from the previous cycle [33]. In Fig. 12, F is plotted as a function of cycle strain for different types of hits: for all hits, hits with amplitude greater than 70 dBAE, and associated hits. The ‘‘all hits’’ data in Fig. 12 show considerable scatter with F < 1 early in the test, increasing to F > 1 with low scatter when ec = 0.3%, and then move downward with increasing scatter thereafter. The other two plots both have much lower scatter and decline to F < 1 when ec > 0.6%, coinciding with the strain where serious damage begins. These data are consistent with the interpretation that, for the most part, the unassociated hits are caused by events with lesser damage. As F is less than unity when ec < 0.6%, the conventional ‘‘all hits’’ analysis shown in Fig. 12 gives misleading results, as it suggests significant and intolerable damage when in fact there was none.

K.W. Campbell, P.H. Mott / Composites Science and Technology 95 (2014) 21–28

27

reloading. This is the reloading hit strength, which is a measure of damage. The mean reloading hit strength was determined for each unloading-reload cycle, and is plotted as a function of the cycle strain ec in the inset (error bars are the standard deviation). The plot shows that reloading hit strength is initially quite small, ca. 7 lV-s, which then jumps very dramatically to 1965 lV-s (an increase of 2.4 orders of magnitude), when ec > 0.6%.

1.2

1.0

F 0.8 5. Conclusions 0.6

All Hits Associated Hits Amplitude > 70 dBAE

0.4 0.2

0.4

0.6

0.8

1.0

c (%) Fig. 12. Felicity ratio F, for the indicated data, as a function of the cycle strain ec. For intelligibility the standard deviation of the ‘‘all hits’’ data are represented by the shaded region; for the other two curves the standard deviation is shown with error bars.

The appraisal fails because it gives equal weight to all events, while clearly some events are more significant than others. This distinction can be accounted for by substituting the cumulative hit strength RMc for the number of hits on the ordinate, as a function of sample strain, as displayed in Fig. 13. The two curves in the main figure are individual samples, showing the minimum and maximum measurements of all tests. The hits that occur during the unload-reloading portions of the cycle are identified by where the data jump leftward; the difference in RMc after reloading (i.e., the vertical distance), to the same strain after reloading, identifies the strength of the hits that occurred while unloading and

Reloading Hit Strength (μV-s)

10

1. The boundary between tolerable and intolerable damage, based on mechanical measurements, occurred at about 0.6% strain. 2. AE data had a bimodal distribution, consistent with two types of damage. By comparison to previous studies, the types are identified as matrix cracking and fiber failure. Matrix cracking predominated during the tolerable damage, while both types of injury occurred during the intolerable damage. 3. The line between tolerable and intolerable damage was related to different AE parameters, such as total number of hits, hit amplitude, duration, and hit strength. Of these, hit strength (i.e., MARSE) gave the clearest indication that the tolerable threshold had been exceeded. 4. Within experimental scatter, there was a linear decrease in stiffness vs. total log(number of hits) and vs. log(cumulative hit strength). This relationship may be used to ensure the damage does not exceed the tolerable limit of a straining component. 5. Some damage – both tolerable and intolerable – occurs while unloading. 6. The traditional felicity plot did not identify the strain when damage becomes intolerable. However, limiting this analysis to high amplitude or associated hits did provide a good correlation. An analog to this analysis using hit strength also showed good correlation.

4

Σ Mc

Acknowledgement

(μV-s)

10

3

10

2

10

1

We are indebted to C.L. Cartwright for his suggestions and comments. We are grateful for Goodrich EPP for the preparation of the samples and for the composite test results. This research was supported by the Naval Sea Command. References 5

10

0.4

0.6

0.8

c (%)

reload Reloading Hit Strength unload

0.4

0.6

0.8

1.0

STRAIN (%) Fig. 13. Modified felicity analysis, showing the cumulative hit strength RMc (substituted for number of hits) as a function of sample strain. Shown are the minimum (left, in blue) and maximum (right, in black) measurements, from all samples. Inset: Cycle reloading hit strength as defined in the figure, as a function of the cycle strain ec. Plotted points are averages over each cycle, and error bars are the standard deviations. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

[1] Davies GAO, Hitchings D, Zhou G. Impact damage and residual strengths of woven fabric glass/polyester laminates. Compos Part A 1996;27(12):1147–56. [2] Guild FJ, Philips MG, Harris B. Acoustic emission studies of damage in GRP. NDT Int 1980;13(5):209–18. [3] Barre S, Benzeggagh ML. On the use of acoustic emission to investigate damage mechanisms in glass-fibre-reinforced polypropylene. Compos Sci Technol 1994;52(3):369–76. [4] Benzeggagh ML, Benmedakhene S. Residual strength of a glass/polypropylene composite material subjected to impact. Compos Sci Technol 1995;55(1):1–11. [5] Zhuang X, Yan X. Investigation of damage mechanism in self-reinforced polyethylene composites by acoustic emission. Compos Sci Technol 2006; 66(3–4):444–9. [6] Benmedakhen S, Kenane M, Benzeggagh ML. Initiation and growth of delamination in glass/epoxy composites subjected to static and dynamic loading by acoustic emission monitoring. Compos Sci Technol 1999;59(2): 201–8. [7] Wooh SC, Daniel IM, Chun HJ. Real-time ultrasonic and acoustic emission monitoring of damage in graphite/epoxy laminates. Compos Eng 1995;5(12): 1403–12. [8] Alander P, Lassila LVJ, Tezvergil A, Vallittu PK. Acoustic emission analysis of fiber-reinforced composite in flexural testing. Dent Mater 2004;20(4):305–12. [9] Johnson M, Gudmundson P. Broad-band recording and characterization of acoustic emission events in composite laminates. Compos Sci Technol 2000; 60(15):2803–18.

28

K.W. Campbell, P.H. Mott / Composites Science and Technology 95 (2014) 21–28

[10] Dzenis YA, Qian J. Analysis of microdamage evolution histories in composites. Int J Solids Struct 2001;38(10–13):1831–54. [11] Huguet S, Godin N, Gaertner R, Salmon L, Villard D. Use of acoustic emission to identify damage modes in glass fibre reinforced polyester. Compos Sci Technol 2002;62(10–11):1433–44. [12] Bocchieri RT, Schapery RA, Gorman MR. Time-dependent microcracking detected in a rubber-toughened carbon-epoxy composite by the modal acoustic emission method. J Compos Mater 2003;37(5):421–51. [13] Loutas TH, Kostopoulos V. Health monitoring of carbon/carbon, woven reinforced composites. Damage Assessment by using advanced signal processing techniques. Part I: Acoustic emission monitoring and damage mechanisms evolution. Compos Sci Technol 2009;69(2):265–72. [14] Yang BH, Zhuang XM, Zhang TH, Yan X. Damage mode identification for the clustering analysis of AE signals in thermoplastic composites. J Nondestruct Eval 2009;28(3–4):163–8. [15] Aggelis DG, Barkoula NM, Matikas TE, Paipetis AS. Acoustic structural health monitoring of composite materials: damage identification and evaluation in cross ply laminates using acoustic emission and ultrasonics. Compos Sci Technol 2012;72(10):1127–33. [16] Aggelis DG, Barkoula NM, Matikas TE, Paipetis AS. Acoustic emission as a tool for damage identification and characterization in glass reinforced cross ply laminates. Appl Compos Mater 2013;20(4):489–503. [17] Pollock AA. Acoustic Emission Inspection. Metals Handbook, 9th ed., vol. 17, ASM, International; 1989. p. 278–94. [18] Hamstad MA. A review: acoustic emission, a tool for composite-material studies. Exp Mech 1986;26(1):7–13. [19] Powers BM, Vinson JR, Hall IW, Hubbard RF. High strain rate properties of cycom 5920/1583 cloth glass/epoxy composites. AIAA J 1997;35(3):553–6. [20] Erhard G. Designing with plastics. Munich: Hanser Publishers; 2006. [21] Mott PH, Smallhorn JM, Campbell KW, Schrader HL, Cartwright CL, Finck DG. Comparison of buckling of arched, rubber core sandwich composite panel to finite element modeling. Compos Part A Appl S 2011;42(7):843–8. [22] Goodrich Corporation EPP. Laboratory service request 0528405 (unpublished); 2005. [23] Ivanov DS, Lomov SV, Bogdanovich AE, Karahan M, Verpoest I. A comparative study of tensile properties of non-crimp 3D orthogonal weave and multi-layer

[24] [25] [26]

[27] [28]

[29]

[30]

[31] [32] [33] [34]

[35]

[36]

plain weave E-glass composites. Part 2: Comprehensive experimental results. Compos Part A Appl S 2009;40(8):1144–57. Kumosa M, Hull D, Price JN. Acoustic emission for stress corrosion cracks in aligned GRP. J Mater Sci 1987;22(1):331–6. Bohse J. Acoustic emission characteristics of micro-failure processes in polymer blends and composites. Compos Sci Technol 2000;60(8):1213–26. Liu PF, Chu JK, Liu YK, Zheng JY. A study on the failure mechanisms of carbon fiber/epoxy composite laminates using acoustic emission. Mater Des 2012;37: 228–35. Argon AS, Cohen RE. Toughenability of polymers. Polymer 2003;44(19): 6013–32. Varadan VK, Varadan VV, Pao Y-H. Multiple scattering of elastic waves by cylinders of arbitrary cross section. I. SH waves. J Acoust Soc Am 1978;63(5): 1310–9. Kerwin EM, Unger EE. Requirements imposed on polymeric materials by structural damping applications. In: Corsaro RD, Sperling LH, editors. Sound and vibration damping with polymers. Washington: American Chemical Society; 1990. p. 317–45. Kaiser J. Erkenntnisse und Folgerungen aus der Messung von Geräuschen bei Zugbeanspruchung von metallischen Werkstoffen. Arch Eisenhüttenwes 1953;24(1–2):43–5. Palmer IG. Acoustic emission measurements on a reactor pressure vessel steel. Mater Sci Eng 1972;11(4):227–36. Harris DO, Dunegan HL. Acoustic emission. 5. Applications of acoustic emission to industrial problems. Non-destruct Test 1974;7(3):137–44. Fuwa M, Harris B, Bunsell AR. Acoustic emission during cycling loading of carbon-fibre-reinforced plastics. J Phys D Appl Phys 1975;8(13):1460–71. Kim S-T, Lee Y-T. Characteristics of damage and fracture process of carbon fiber reinforced plastic under loading-unloading test by using AE method. Mater Sci Eng 1997;A234–236:322–6. Lachaud F, Lorrain B, Michel L, Barriol R. Experimental and numerical study of delamination caused by local buckling of thermoplastic and thermoset composites. Compos Sci Technol 1998;58(5):727–33. Miller RK, McIntire P, editors. Acoustic emission testing. Nondestructive testing handbook, 2nd ed., vol. 5. Columbus (OH): American Society for Nondestructive Testing; 1987.