Identifying the breaks of wool fibers based on the waveform analysis of acoustic emission

Abstract

Acoustic emission (AE) of fiber break is the sound waves reflecting the stress waves of material during rupturing. It can be employed to evaluate the tensile properties of single fiber during the fiber-bundle tensile test. The key to this technique is to identify AEs of fibers at the break point accurately. However, the severe interference introduced by the background noise and the inter-fiber friction is inevitable. In this paper, a filtering method based on the wavelet de-noising is used to remove the background noise. In order to identify the fiber-broken AEs, a combined parameter that characterizes the amplitude and the density fluctuation along the time axis is proposed. Multiple test results indicate that this parameter can be used to identify wool fiber-broken AEs with accuracy rates higher than 95%.

Keywords

acoustic emission fiber break wavelet de-noising waveform parameter analysis

The tensile properties of fibers are vital characteristics that determine both the performance and the price of textile products made from it. Due to the variety of fibers, it is necessary to test a large number of fibers to obtain their average performance. Giving the theoretical^1–5 relationships of the tensile properties between the single fiber and the fiber bundle, the tensile properties of a single fiber can be estimated from the tensile behavior of the fiber bundle. Therefore, the fiber bundle tensile test is used to evaluate the fiber tensile properties for its convenience and efficiency.^6,7 Technically, the fiber-bundle tensile test is based on a parallel set of fibers, which are usually tightly arranged.

As an effort to investigate the strength distribution of single fibers during the fiber-bundle tensile test, the acoustic emission (AE) method is introduced for synchronous measurement. It is a nondestructive method for damage evaluation during mechanical loading of materials.^8,9 The technique is used, for example, to study the damage mechanisms in composite materials, that is, the structure-borne AEs during the formation and development of cracks are registered and correlated with the destruction process.^10–12 In the context of this paper, the recorded AEs and the corresponding spectrums can be regarded as the “snapshot” of the deformation and the destruction of the fiber bundle along the time axis. Since it has been proved that the break of each fiber can be recorded in the fiber-bundle tensile test,¹³ indicating each fiber breaks one by one, all fiber-broken AEs can be obtained.

Since the inter-fiber friction AEs from the frictions among fibers can be easily mistaken for fiber-broken AEs, it is important to identify the fiber-broken AEs accurately. The conventional method is to set a threshold between the amplitudes of the fiber-broken AEs and those of the inter-fiber friction AEs.^14,15 The AEs of higher amplitude are ascribed to fiber breaks, while the AEs of lower amplitude are ascribed to inter-fiber friction.

Although the AE method has been adopted in determining the tensile properties of man-made fibers,^16–21 it has not been widely used to test natural fibers. Only a few papers^22,23 have been published concerning the AEs generated by cotton fiber breaks, where the methods of identifying the fiber-broken AEs were not fully investigated. Compared to man-made fibers, the difficulty of identifying the fiber-broken AEs of natural fibers is due to two possible reasons. Firstly, the surfaces of natural fibers are relatively coarser than those of man-made fibers due to their apparent features (joint spots on the ramie fiber, natural convolution on the cotton fiber, and the natural scales on the wool fiber). These features may introduce more inter-fiber friction AEs with high amplitudes. Secondly, the lubricant such as petroleum oil that is introduced to minimize the inter-fiber friction among man-made fibers^24,25 is not suitable for natural fibers, since the penetration of the lubricant will affect their tensile properties. Regarding these limitations, we believe more efforts should be focused on the identification of the fiber-broken AEs for natural fibers.

In this paper, both single fibers and fiber bundles of dry wools are used for probing the suitable means of AE identification. A new method based on wavelet de-noising and waveform parameter analysis is proposed to identify the fiber-broken AEs in the fiber-bundle tensile test. In the second section, we explain signal acquisition and processing. In the third section, we depict and discuss the experimental results. We conclude our work in the final section.

Experimentation

Testing device

A computer-based instrument (InFiBTensor) for AE acquisition has been developed for this investigation. It consists of two main parts, a tensile strength tester and an acoustic detector equipped with a microphone and an audio card for data collection. The microphone with a frequency response from 50 to 20,000 Hz can detect AEs from the breaks of single fibers. The sampling rate of the audio card is 16 bit and 44.1 kHz. The schematic diagram of the testing device is illustrated in Figure 1.

Figure 1.

Schematic diagram of the testing device.

Sample preparation and measurement

According to our observation, the finer the wools are, the harder it will be to identify the fiber-broken AEs. In this paper, wool fibers with fineness around 18 and 32 µm collected from various top manufacturers are used to validate our proposed method.

The samples for the single-fiber and the fiber-bundle tensile tests are prepared and conditioned under the standard conditions of 20 ± 2℃ and 65 ± 3% relative humidity (RH). In order to reduce the floating fibers and to install the microphone, the clamping length is set to10 mm, which is much shorter than the average length of wools. The microphone is positioned 5 mm²² from the fiber.²² The extension rate is set at 3 mm/min to avoid the noise introduced by the movement of the stepping motor. This is actually an optimized setting after testing various loading rates and gauge lengths on InFiBTensor.

During the breaking procedure, the fracture of each fiber produces an AE. The microphone detects the AEs simultaneously, and the signals from the load cell are recorded. The raw data are sent to a computer for registering and processing. The output is an AE spectrum combined with a full stretching diagram, as shown in Figure 2.

Figure 2.

Acoustic emission spectrum combined with full stretching diagram.

Original acoustic signal

Since the sound is not collected in an anechoic chamber, the original acoustic signals obtained from the fiber-bundle tensile test are composed of three parts: the background noise, the fiber-broken AEs, and the inter-fiber friction AEs. According to their waveforms, these three types of AEs can also be categorized as continuous AEs and burst AEs, as shown in Figure 3 and Table 1.

Figure 3.

Different kinds of acoustic emission (AE) obtained from the fiber-bundle tensile test.

Table 1.

Classification of acoustic emissions (AEs)

Category	Attribute
Continuous AEs	Background noise
Burst AEs	Inter-fiber friction AEs Fiber-broken AEs

As shown in Figure 3, the amplitudes of the burst AEs (i.e., the inter-fiber friction AE and the fiber-broken AE) are similar. The differences between their amplitudes are not distinct enough. Both of them fluctuate with the background noise and parts of them are mixed up. Obviously, to measure the waveforms of the burst AEs accurately, the background noise should be removed first. In this paper, this is performed by wavelet de-noising.

Wavelet de-noising

De-noising is one of the widely discussed techniques in signal processing. Due to its flexibility in choosing the wavelet basis in wavelet transform and the characteristic of time-frequency localization, the wavelet de-noising has the capacity to preserve the frequency contents of the burst AEs and to remove the frequency contents of the background noise.^26,27

The observations of the original signal are given by

Y_{i} = F_{i} + Z_{i}

(1)

where F_i is the signal. Z_i is the noise satisfied with

N (0, 1)

distribution,

i = l, 2 \dots N

, where N is the length of the data. The goal is to recover F_i from the noisy data Y_i with a small error.

Denoting $Y_{j, k}$ as the wavelet coefficients of Y_i, the transform of Equation (1) to the wavelet domain is

Y_{j, k} = W_{j, k} + Z_{j, k}

(2)

where

j = l, 2 \dots N - 1

denotes the decomposition level.

K = 0, 1 \dots 2 j - 1 - 1

is the index of the coefficient in this level.

w_{j, k}

is the wavelet coefficient of F_i and

Z_{j, k}

satisfies with

N (0, 1)

distribution.

The coefficients of the wavelet transform usually are sparse. The coefficients with small magnitude can be considered as pure noise and should be set to zero. Therefore, significant coefficients can be extracted by setting the coefficients whose absolute value is below a certain threshold (denoted as λ). The thresholded wavelet coefficients are obtained using either hard or soft thresholding functions, $T_{hard}$ and $T_{soft}$ , expressed as follows:

T_{hard} = {Y_{j, k} - λ | Y_{j, k} | \geq λ 0 | Y_{j, k} | < λ

(3)

T_{soft} = {Y_{j, k} - λ Y_{j, k} \geq λ Y_{j, k} + λ Y_{j, k} \leq - λ 0 | Y_{j, k} | < λ

(4)

Now reconstructing the original sequence from the thresholded wavelet coefficients will lead to a de-noised (smoothed) version of the original sequence.

In our practice, the detailed steps for the wavelet de-noising are as follows: (1)

decompose the original signal to six-level with wavelet basis db5 to produce the noisy wavelet coefficients;

(2)

perform algorithm “minimax”²⁸ to calculate λ:

λ_{minimax} = {σ_{n} (0.3936 + 0.1829 \log_{2} n), n > 32 0 n < 32

(5)

where n is the value of

W_{j, k}

and

σ_{n}

is the variance of Z_i;

(3)

select the soft thresholding function to remove noises.

(4)

perform inverse wavelet transforms of the thresholded wavelet coefficients to obtain a de-noised signal.

As shown in Figure 4, a section of the original signal is used as an example to demonstrate the de-noised effect. In Figures 4(a) and (b), the background noises are almost fully removed, and the details of the burst AEs in Figure 4(b) are well preserved, as shown in 4(d).

Figure 4.

The effect of wavelet de-noising: (a) original signal; (b) amplification of the rectangle zone of (a); (c) de-noised signal; (d) zooming in the rectangle zone of (c).

Extraction of waveform parameters

After removing the background noises, the next challenge is to determine which part of the remaining burst AEs are caused by the fiber breaks. Since the AEs depend on the characteristics of their acoustic sources, this can be regarded as a problem of feature analysis.

Basic parameter

According to our observation, eight basic parameters can be specified in characterizing the AE waveforms, as shown in Figure 5.

Duration (t_D): t_D is the time difference between the first and the last threshold crossings.

Rise time (t_R): t_R is the time interval between the first threshold crossing and the maximum positive peak.

Amplitude (A_max): A_max is the maximum positive peak in a waveform.

Counts(C): C refers to the number of positive peaks whose amplitude is greater than the threshold.

Energy (E): E is the measure of the shaded area.

Ratio of rise time to attenuation time (t_R/t_A): attenuation time (t_A) is the difference between t_D and t_R.

A_pre is the height of the positive peak locating just before the maximum peak.

t_P is the time difference between adjacent positive peaks.

Figure 5.

Illustration of the waveform and the parameters.

A threshold is used to determine the waveform parameters. Only the peaks higher than this threshold would be measured. The value of this threshold can be determined by first measuring the A_pre of all burst AEs, and then setting it as 0.5MIN (A_pre). Since the A_pre shown in Figure 5 is not the minimum in our test, the threshold is not equal to the 0.5 A_pre.

Structure parameter

Hundreds of single fiber tensile tests have been carried out to investigate the shape features of the fiber-broken AE. The typical waveform of the wool fiber-broken AE is shown in Figure 6 (after de-noising). Since the target of this paper is to establish a way to identify the fiber-broken AEs from various inter-friction AEs, multiple bundle tensile tests have also been conducted. As shown in Figure 7 (after de-noising), except for the waveform (f) that is similar to the one shown in Figure 6, all other waveforms belong to the inter-friction AEs.

Figure 6.

Typical waveform of the wool fiber-broken acoustic emission.

Figure 7.

The typical acoustic emission waveforms in the fiber-bundle tensile test.

The eight parameters proposed in the Extraction of waveform parameters section are basic measures to describe the shape of the waveform after the wavelet de-noising. Comparing Figure 7 to Figure 6, an efficient measure needs to be designed to describe the difference between the fiber-broken AEs and the inter-fiber frictions AEs. Since the fiber-broken AE is generated from the rupture procedure of the fiber, it should be focused on the maximum peak (A_max) and its neighborhood along the time axis. Based on this assumption, two structure parameters are proposed to describe the “structure” of the waveform.

As shown in Table 2,

A_{cri}

and

t_{Pcri}

are the critical values of the structure parameters,

A_{max} / A_{pre}

and

t_{P min} / t_{P max}

, respectively. In the following section, we will depict how to use these structure parameters to deduct a better solution in the AE identification.

Table 2.

Structure parameters

Parameter	Method for identifying fiber-broken AEs
$A_{max} / A_{pre}$	$A_{max} / A_{pre} > A_{cri}$
$t_{Pmin} / t_{Pmax}$	$t_{Pmin} / t_{Pmax} > t_{Pcri}$

AE: acoustic emission.

Results and discussion

In this section, we conduct both single-fiber and fiber-bundle tensile tests to find the suitable parameter that can be employed in the AE identification.

Statistical analysis of the basic parameters

Single-fiber tensile test: basic

The single-fiber tensile test is designed to avoid the influence of the inter-fiber friction. We can obtain the pure fiber-broken AEs and analyze the distribution ranges of their waveform parameters through this type of test. Since 200–300 fibers could represent a certain kind of wool,²⁹ 200 fibers are randomly gathered from the wool top for testing. Basic parameters t_D, t_R, A, C, E, and t_R/t_A are extracted from the 200 fiber-broken AEs and are summarized in the histograms respectively, as shown in Figure 8. The statistical analysis is given in Table 3.

Figure 8.

Distribution histogram of the fiber-broken acoustic emissions using the basic parameters.

Table 3.

Distribution statistics of the basic parameters of the fiber-broken acoustic emissions

Index	t_D (s)	t_R (s)	A_max (mV)	C	E	t_R/t_A (%)
Average	0.0017	0.00045	201.04	16	1772.15	41.19
Median	0.0018	0.00043	185.22	15	1557.60	37.50
Minimum(ɛ)	0.0005	0.00023	50.684	6	257.42	16.67
Maximum(δ)	0.0033	0.00077	760.83	35	6701.60	80.56
SD	0.0005	0.00011	104.96	7	1020.85	16.55

As shown in Figure 8, these data fit into skewed distribution rather than normal distribution. According to Table 3, the median can be used as a measure of central tendency of these distributions. The minimum ɛ (ɛ = MIN{p}, p = t_D, t_R, A_max, C, E, t_R/t_A) and the maximum δ (δ = MAX{p}, p = t_D, t_R, A_max, C, E, t_R/t_A) can be used to define the ranges of them as statistical regularity. Since 200 fibers are sufficient to represent the wool under testing, we take ɛ and δ as the distribution range of the fiber-broken AEs of the correspondent parameter. If the parameter of an AE locates beyond its distribution range, it then will be regarded as the inter-fiber friction AE.

Fiber-bundle tensile test: basic

In order to identify the fiber-broken AEs and the inter-fiber friction AEs with ɛ and δ, the fiber-bundle tensile test is carried out. A wool top is combed to remove twists and floating fibers. About 200–300 fibers are randomly selected from this top and are tightly arranged to form a parallel fiber bundle. The actual number of broken fibers can be counted after the test.

To explore the features of the AE waveforms, six groups of fiber-bundle tensile tests are conducted; the actual numbers of the broken fibers are given in Table 8 (18 µm). For the convenience of explanation, we use No. 6 as an example. We manually counted 263 broken fibers after the fiber-bundle tensile test, while our proposed system recorded 386 burst AEs. The basic parameters t_D, t_R, A_max, C, E, and t_R/t_A are extracted from these 386 burst AEs and are summarized in the histograms, as shown in Figure 9.

Figure 9.

Distribution histogram of the break-broken and the inter-fiber friction acoustic emissions using the basic parameters.

Figure 10.

Distribution histogram of the fiber-broken acoustic emissions using the structure parameters.

Figure 11.

Histogram of the fiber-broken acoustic emissions (AEs) and the inter-fiber friction AEs using the structure parameters.

According to Figure 9, all the distributions are beyond ɛ and δ shown in Table 3. To display this more clearly, slash bars are used to mark the distributions that are less than ɛ. Black bars are used to mark the distributions that are greater than δ. Since these two kinds of distributions do not comply with the statistical regularity of the fiber-broken AEs, they are mainly caused by the inter-fiber friction AEs, while the unmarked bars are mainly caused by the fiber-broken AEs.

Analysis of the accuracy rate

As shown in Figure 9, the number of the fiber-broken AEs can be counted using different parameters. The accuracy rates of different parameters can be calculated by dividing the numbers of the fiber-broken AEs by the actual number of the broken fibers, as shown in Table 4.

Table 4.

Accuracy rates of the basic parameters

Parameter	t_D (s)	t_R (s)	A_max (mV)	C	E	t_R/t_A (%)
Accuracy rate	81.4%	88.6%	90.9%	84.0%	87.1%	79.9%

Statistical analysis of the structure parameters

Since the structure parameters $t_{P min} / t_{P max}$ and $A_{max} / A_{pre}$ are related with the structure characters of the waveforms, as shown in Table 2, they are also used to analyze the same experiment data with the same methods depicted in the Single-fiber tensile test: basic and Fiber-bundle tensile test: basic sections.

The results of single-fiber tensile test (structure) are shown in Figure 10 and Table 5. The results of fiber-bundle tensile test (structure) are shown in Figure 11 and Table 6. Compared to Table 4, it can be seen that all the accuracy rates of the structure parameters are higher than those of the basic parameters. This indicates that the structure parameters are more suitable for identifying the fiber-broken AEs than the basic parameters. Although all the proposed parameters can be employed in identifying the fiber-broken AEs, there are still overlaps between the distributions of the fiber-broken AEs and the inter-fiber friction AEs. In order to pursue higher accuracy and reliability, a combined parameter is designed, as shown in Figure 12.

Figure 12.

Schematic diagram of the function of the combined parameter: (a) the result using the basic or structure parameter; (b) the result using the combined parameter.

Table 5.

Distribution statistics of the structure parameters of the fiber-broken acoustic emissions

Parameter	Average	Median	Minimum(ɛ)	Maximum(δ)	SD
$t_{Pmin} / t_{Pmax}$	5.3	5.0	1.7	9.1	1.9
$A_{max} / A_{pre}$	69.1	26.1	5.8	1098.3	144.9

Table 6.

Accuracy rates of the structure parameters

Parameter	Accuracy rate
$t_{Pmin} / t_{Pmax}$	93.2%
$A_{max} / A_{pre}$	92.8%

Combined parameters and accuracy rates

Selection of the combined parameters

In order to minimize the overlaps shown in Figure 12(a), the design of the combined parameter should reduce the distribution of the inter-fiber friction AEs and should enlarge the distribution of the fiber-broken AEs. Since both the basic parameters and the structure parameters can be employed for constructing the combined parameter, we set up two rules for screening: (a) the parameter with good capacity in identifying the fiber-broken AEs and the inter-fiber friction AEs should be selected; (b) the parameter with low correlation with each other should be taken. These are to ensure the independence of each parameter and to rearrange the AEs within the overlaps more effectively.

As shown in Table 7, the absolute value of the correlation coefficients among

t_{R} / t_{A}

A_{max} / A_{pre}

and

t_{P min} / t_{P max}

are much lower than those of other parameters. These parameters also have superior capacity in identifying the fiber-broken AEs and the inter-fiber friction AEs according to Tables 4 and 6. Therefore, we select them to construct the combined parameter. Since the correlation coefficients between

A_{max} / A_{pre}

and

t_{P min} / t_{P max}

are positive, and the correlation coefficients between

t_{R} / t_{A}

and the others are negative, we can use the multiplication of

A_{max} / A_{pre}

t_{P min} / t_{P max}

and

\frac{1}{t_{R} / t_{A}}

to reduce the overlaps between the fiber-broken AEs and the inter-fiber friction AEs.

Table 7.

The correlation coefficients between the different parameters

Parameter	t _D	t _R	A	C	E	$t_{R} / t_{A}$	$A_{max} / A_{pre}$	$t_{Pmin} / t_{Pmax}$
t _D	1.000
t _R	0.793	1.000
A	0.821	0.603	1.000
C	0.948	0.656	0.855	1.000
E	0.861	0.638	0.944	0.875	1.000
$t_{R} / t_{A}$	–0.214	0.253	–0.273	–0.312	–0.268	1.000
$A_{max} / A_{pre}$	0.204	0.109	0.137	0.222	0.152	–0.082	1.000
$t_{Pmin} / t_{Pmax}$	0.738	0.776	0.600	0.635	0.604	–0.054	0.140	1.000

The combined parameter X is designed as

X = \frac{\frac{A_{max}}{A_{pre}} \times (\frac{t_{Pmin}}{t_{Pmax}}) 2}{(\frac{t_{R}}{t_{A}}) 2} = \frac{A_{max} (\frac{t_{D} - t_{R}}{t_{Pmax}}) 2}{A_{pre} (\frac{t_{R}}{t_{Pmin}}) 2}

In the expansion of X, $\frac{t_{D} - t_{R}}{t_{Pmax}}$ represents the number of the longest-span peaks that could be contained during the attenuation time. $\frac{t_{R}}{t_{Pmax}}$ represents the number of the shortest-span peaks that could be contained during the rising time. In this sense, the combined parameter X characterizes the amplitude and the density fluctuation of its neighborhood along the time axis. To investigate if applying various exponentiations for ( $t_{Pmin} / t_{Pmax}$ ) and ( $t_{R} / t_{A}$ ) can further magnify the difference between the fiber-broken AEs and the inter-fiber friction AEs, multiple exponentiations have been tried. As shown in Figure 13, the square processing is proved to have the best effect.

Figure 13.

Comparison of the identifying effects using the different values of power for $t_{Pmin} / t_{Pmax}$ and $t_{R} / t_{A}$ .

Analysis of the accuracy rate

According to Tables 3 and 5, putting ɛ of $A_{max} / A_{pre}$ and $t_{Pmin} / t_{Pmax}$ , as well as δ of $t_{R} / t_{A}$ into X, the critical values of the combined parameter can be calculated as X_Cri = 14.90. X is also applied to analyze the waveforms with the same experimental data and the same method depicted in the Fiber-bundle tensile test: basic section. The results are shown in Figure 14. Since the distributions of X are too sparse to plot, the logarithmic coordinates are adopted.

Figure 14.

Histogram of the fiber-broken acoustic emissions (AEs) and the inter-fiber friction AEs using the combined parameter X.

In Figure 14, the position marked by the arrow indicates X_Cri. The distributions less than X_Cri are mainly caused by the inter-fiber friction AEs, while the others are mainly caused by the fiber-broken AEs. Therefore, the identified number of the fiber-broken AEs using X is 253. The corresponding accuracy rate is 96.2%. Since the combined parameter X describes the uniqueness of the fiber-broken AEs using amplitude and the density fluctuation of peaks simultaneously, it is more suitable in distinguishing the fiber-broken AEs from the inter-fiber friction AEs.

Now our proposed method to identify the fiber-broken AEs for various wool fiber bundles can be summarized as follows.

Step 1: Conduct 200 single-fiber tensile tests and calculate X_Cri of the tested wool.

Step 2: Conduct the fiber-bundle tensile test and calculate X for each detected waveform after the wavelet de-noising.

Step 3: Compare X with X_Cri to determine the fiber-broken AEs.

Although we do not use a more sophisticated method such as the neural network to categorize the different types of AEs, the procedure of getting X_Cri through the 200 single-fiber tensile tests can be regarded as a procedure of data training. The result is a promised output with high accuracy. As shown in Table 8, six groups of replication tests with the wool (18 µm) are conducted to examine the performance of our method, and six more groups with the wool (32 µm) as additional verification.

Table 8.

Comparison between the actual and the detective number of wool fibers using the combined parameter X

Wool	18 µm						32 µm
Group	1#	2#	3#	4#	5#	6#	1#	2#	3#	4#	5#	6#
Actual value	208	285	254	231	303	263	233	259	280	243	281	210
Detected value	199	271	244	221	290	253	227	251	273	239	272	205
Accuracy rate/%	95.7	95.1	96.1	95.7	95.7	96.2	97.4	96.9	97.5	98.4	96.8	97.6

In Table 8, with all the accuracy rates higher than 95% and the average accuracy rate at 96.6%, it is believed that the use of the combined parameter X provides a good reliability in the fiber-broken AE identification. Comparing to the wool (18 µm), the accuracy rates of the wool (32 µm) are higher, which indicates that finer wool fibers bring more challenges to the fiber-broken AE identification.

The success of the combined parameter X also proves that the wool fibers break one by one in the fiber-bundle tensile tests. To further verify this, we record the occurrence time of A_max of each fiber-broken AE in the 12 groups of fiber-bundle tensile tests and arrange them in ascending order. Since their conclusions are similar, we still use group No.6 (18 µm) as the example, as shown in Figure 15.

Figure 15.

The time distribution of the fiber-broken acoustic emissions (AEs) in the fiber-bundle tensile test (the horizontal axis represents the sequence of the fiber-broken AEs and the vertical axis represents the occurrence time of A_max).

It can be seen from Figure 15 that the curve increases monotonously. For the 253 fiber-broken AEs, there is no overlap for their occurrence time of A_max. The minimum time interval can be calculated as 0.0046 s, which is long enough to identify the fiber-broken AEs.

According to Table 8, all the detected values are smaller than the actual values. This means some fiber breaks cannot be detected using our proposed method for two possible reasons. Firstly, the fiber-broken AEs of some fine fibers or damaged fibers are too weak to be detected by the microphone. Secondly, some fiber-broken AEs are still mistaken as the inter-friction AEs using the combined parameter X. The future work to improve this can be engaged in two directions. Firstly, more advanced acoustic sensors need to be applied to detect a larger range of frequency, that is, ultrasonic frequency. Secondly, more effective parameters should be further developed.

Conclusion

In this paper, a method based on wavelet de-noising and waveform analysis has been proposed to identify fiber-broken AEs. The wavelet de-noising is used to remove the background noise and to preserve the waveform details of the burst AEs. The waveform analysis indicates that the basic parameters and the structure parameters can only be used for preliminary identification. More accurate identification should be fulfilled via the combined parameters X, with which the accuracy rates are higher than 95%.

Footnotes

Funding

This work was supported by Shanghai Natural Science Foundation (Project No. 14ZR1401100) and “the Fundamental Research Funds for the Central Universities” (11D10122, 13D110111).

References

Daniels

. The maximum of a Gaussian process whose mean path has a maximum, with an application to the strength of bundles of fibres. Adv Appl Probab 1989; 21: 315–333.

Coleman

. On the strength of classical fibres and fibre bundles. J Mech Phys Solid 1958; 7: 60–70.

Nachane

Iyer

. Prediction of bundle strength from single-fiber test data. Textil Res J 1980; 50: 639–641.

Daniels

. The statistical theory of the strength of bundles of threads I. Proc Roy Soc Lond Math Phys Sci 1945; 183: 405–435.

Radhakrishnan

Shelat

. Tensile behavior of cut skeins and single threads. Textil Res J 1961; 31: 1063–1069.

Vas

Rácz

. Modeling and testing the fracture process of impregnated carbon-fiber roving specimens during bending: Part I–Fiber bundle model. J Compos Mater 2004; 38: 1757–1785.

Frydrych

. Relation of single fiber and bundle strengths of cotton. Textil Res J 1995; 65: 513–521.

Beattie

. Acoustic emission, principles and instrumentation, Albuquerque, NM: Sandia National Labs, 1983.

Eitzen

Wadley

. Acoustic emission: establishing the fundamentals. J Res Natl Bur Stand 1984; 89: 75–100.

10.

Giordano

Calabro

Esposito

. An acoustic-emission characterization of the failure modes in polymer-composite materials. Compos Sci Tech 1998; 58: 1923–1928.

11.

Narisawa

Oba

. An evaluation of acoustic emission from fibre-reinforced composites. J Mater Sci 1984; 19: 1777–1786.

12.

Marec

Thomas

J-H

El Guerjouma

. Damage characterization of polymer-based composite materials: Multivariable analysis and wavelet transform for clustering acoustic emission data. Mech Syst Signal Process 2008; 22: 1441–1464.

13.

Hamstad

Moore

. Acoustic emission from single and multiple Kevlar 49 filament breaks. J Compos Mater 1986; 20: 46–66.

14.

R’Mili

Moevus

Godin

. Statistical fracture of E-glass fibres using a bundle tensile test and acoustic emission monitoring. Compos Sci Tech 2008; 68: 1800–1808.

15.

Jihan

Siddiqui

Sweet

. Fracture strength of E-glass fibre strands using acoustic emission. NDT & E Int 1997; 30: 383–388.

16.

Cowking

Attou

Siddiqui

. Testing E-glass fibre bundles using acoustic emission. J Mater Sci 1991; 26: 1301–1310.

17.

Okoroafor

Hill

. Investigation of complex failure modes in fibre bundles during dynamic mechanical testing using acoustic emission and Weibull statistics. J Mater Sci 1995; 30: 4233–4243.

18.

Wilson

. Statistical tensile strength of nextelTM 610 and nextelTM 720 fibres. J Mater Sci 1997; 32: 2535–2542.

19.

Lissart

Lamon

. Statistical analysis of failure of SiC fibres in the presence of bimodal flaw populations. J Mater Sci 1997; 32: 6107–6117.

20.

Oskouei

Ahmadi

. Fracture strength distribution in E-glass fiber using acoustic emission. J Compos Mater 2010; 44: 693–705.

21.

Perepelkin

. New impulse-acoustic method for testing and estimating the transversal unevenness and defectiveness of filaments in chemical yarns. Fibre Textil East Eur 2007; 15: 36–36.

22.

Lyons

Graham

. Acoustic measurement of fiber properties. Textil Res J 1975; 45: 549–553.

23.

Velu

Ellison

Dai

. Single fiber acoustic emission and tensile properties from cotton bundle fracture tests. Textil Res J 1999; 69: 616–620.

24.

Hill

Okoroafor

. Weibull statistics of fibre bundle failure using mechanical and acoustic emission testing: the influence of interfibre friction. Composites 1995; 26: 699–705.

25.

Phoenix

. Statistics for the time-dependent failure of Kevlar-49/epoxy composites: Micromechanical modeling and data interpretation, Livermore, CA: CA: Lawrence Livermore National Lab., 1983.

26.

Donoho

. De-noising by soft-thresholding. IEEE Trans Inform Theor 1995; 41: 613–627.

27.

Mallat

Hwang

. Singularity detection and processing with wavelets. IEEE Trans Inform Theor 1992; 38: 617–643.

28.

Donoho

Johnstone

. Minimax estimation via wavelet shrinkage. Ann Stat 1998; 26: 879–921.

29.

Chu

. Textile physics, 2nd ed. Shanghai: Donghua University Press, 2002, pp. 315–318.