Fiber recognition with machine learning methods by fiber tensile fracture via acoustic emission method

Abstract

Energy release usually accompanies the single-fiber tensile fracture, and can be monitored using acoustic emission technology. Generated during the process of molecular structure fracture of various fibers, the acoustic emission signals can be extracted to identify different fracture types of fiber, which is especially important to the yarn formation process. In this study, a low-noise fiber-stretching device was employed to process the weak-intensity signal generated during fiber tensile fracture; in addition, the Hilbert–Huang transform (HHT), principal component analysis (PCA) and least squares support vector machine (LSSVM) algorithms were combined to identify the collected acoustic emission signals of polyester and cotton fibers. At the same time, it was verified that compared with the single-fiber breaking acoustic emission signal obtained by the electronic single-fiber strength tester, the signal acquisition device based on pneumatic components proposed in this paper can significantly improve the signal-to-noise ratio of the signal. According to the algorithm recognition results, the recognition rate of the two fibers increased from 74% to 95%.The experimental results indicate successful measurements of different fractures of two types of fiber.

Keywords

fiber tensile fracture acoustic emission feature extraction spectrum analysis

As an important component of textile, the yarn, whose performance has tremendous influence on the application of its finished products in a direct way, should be specially focused on. At present, high-performance and multi-functional fibers, which not only meet people’s daily needs but are also important in many industries, continue to be developed. Generally speaking, blended yarns composed of multiple fibers are the most commonly used yarns, and different fiber materials will present different distributions as the yarn structure changes. Aimed at evaluating fiber properties, traditional methods mainly concentrate on mechanical properties, such as initial modulus, breaking strength, breaking elongation, etc. On one hand, researchers have made use of the acoustic emission (AE) technology to study the vibration signal generated by the fiber tensile fracture; on the other hand, they have introduced artificial intelligence technology into the field of textile detection, realizing the automatic recognition of different kinds of signal generated by the fiber tensile fracture via pattern recognition methods.

Considered as an exceedingly common physical phenomenon in the nature, AE occurs when a material undergoes irreversible change in its internal structure. The energy accumulated inside the material will be released and transmitted to the surrounding environment in the form of an elastic wave during this instability process, and this can be recorded by a sensor. AE is a remarkably versatile and non-invasive way to collect information about a material or structure. It has been proven by practical application that AE technology is extremely useful in inspecting and monitoring pipelines, storage tanks, bridges, aircraft, bucket trucks and a variety of composite and ceramic components.^1–5 In actual applications, AE technology refers to the detection and diagnosis of the target through the reception and analysis of the AE signal. The detection, recording and analysis of the signal are performed using certain equipment and setups.^6–8 AE technology plays a distinctly important role in material destructive testing and evaluation, owing to its advantages of convenient operation, dynamic real-time detection, and high sensitivity and resolution.^9–12

The AE signal generated by the fiber tensile fracture is, in essence, the release of a transient elastic stress wave in the process of deformation or fracture of the fiber material under the action of external force. The transient elastic stress wave is received and then converted into an AE signal by the sensor, through the transmission channel. Different from traditional signal-processing methods involved in analysis and transformation of tested signals for useful feature extraction, the AE signal-processing method includes Fourier transform, wavelet analysis, blind source separation, Hilbert–Huang transform (HHT) and high-order statistical analysis. Proposed by Huang,¹³ HHT is a novel signal analysis method whose core idea lies in the identification of the signal which is used as the basic signal and is composed of intrinsic mode function (IMF). The HHT method exhibits preferable adaptability in processing the AE signal generated during the process of fiber tensile fracture, which is characterized by non-linear and non-stationary features. It can be divided into two steps when executing the analysis on the target signal: empirical modal decomposition (EMD) and HHT. Among these, the EMD method is susceptible to modal aliasing, which makes false components appear in decomposition results. With the objective of overcoming the shortcomings of modal aliasing, Huang proposed the ensemble empirical modal decomposition (EEMD), which combines white Gaussian noise with the original signal with uniform power and zero mean value so as to complete data reconstruction. The noise in the signal refers to the combination of white Gaussian noise, which enables the non-stationary signal to be distributed in a continuous and even way. Because HHT does not use a predetermined basis function for signal component extraction, HHT can accurately obtain the frequency and instantaneous amplitude of the signal. Huang’s research pointed out the mathematical meaning and limitations of traditional time–frequency signal-processing methods (such as fast Fourier transform and wavelet transform), and revealed the advantages of HHT in time–frequency signal processing, especially for non-linear and non-stationary signal-processing advantages.¹⁴ In non-linear and non-stationary signal processing, the instantaneous frequency and amplitude extracted by HHT during time–frequency decomposition have more practical physical significance. HHT was used to study non-linear and irregular water waves, and the results showed that the values of instantaneous frequency and amplitude in the Hilbert spectrum are more correlated with water wave parameters.¹⁵

In the process of the fiber tensile fracture, a rather weak vibration signal will be received by the sensor because of the generated low transient elastic stress wave. Furthermore, the noise interference caused by different sources will result in aliasing distortion of the signal; in addition, the real information will be annihilated by the noise signal during feature extraction from the signal. Taking these points into consideration, it is difficult to collect the AE signal generated during the fiber tensile fracture process. Consequently, it is of utmost importance to improve the signal-to-noise ratio (SNR) of the AE signal in the fiber tensile fracture testing process, especially for the weak signal. Traditional signal analysis methods usually consist of the following steps: adaptive noise cancellation, digital filtering and discrete statistical averaging, which are generally used for signal preprocessing.¹⁶ Guo et al. proposed an adaptive noise elimination method based on the EMD method for water leakage detection and positioning of water pipelines, which improved the SNR to 16 db.¹⁷ At the same time, some researchers put forward the adaptive wavelet technique to perform the noise reduction on the signal, which made up for the shortcomings of the continuous wavelet in identifying gear mesh faults and improved the accuracy of fault identification.¹⁸ Although the noise can be reduced to a certain extent and the transient component of the signal can be smoothed, the true information of the signal still fails to be effectively identified. Produced by the fiber tensile fracture, the AE signal is prone to be influenced by noise during the interfered acquisition process. The noise intensity is so high that it may overwhelm the information carried by the effective signal, affecting the analysis on the subsequent effective signal and feature extraction to some extent. In order to improve the SNR of the AE signal generated by the fiber tensile fracture, a pneumatic stretching device was comprehensively designed and appropriately utilized to obtain uniform stretching until fiber fracture through controlling the airflow. Compared with the electronic single-fiber strength tester driven by stepper motor, the intensity of noise generated by this aforesaid device during the signal acquisition process can be significantly reduced to guarantee preferable purity of signal.

In this paper, HHT was used to analyze the AE signal generated by the fiber tensile fracture, followed by the extraction of time–frequency–energy feature from the signal. Since the AE signal generated by the fiber tensile fracture presents a large feature dimension, the principal components analysis (PCA) was adopted for linear transformation of feature data to obtain multiple component sets with different dimensions, so as to realize dimensionless processing of data from high to low dimension and then optimize the feature components according to experimental requirements. Moreover, a classifier was established by least squares support vector machine (LSSVM), and the recognition of different signals generated by the fiber tensile fracture was also completed according to relevant extracted features. Compared with traditional neural networks, the support vector machine (SVM) possesses prominent advantages: (1) a strong theoretical background provides the SVM with a high generalization capability so that the local minimum can be avoided; (2) the SVM has such a feature that the solution can be quickly obtained via a standard algorithm (quadratic programming); (3) there is no need for the SVM to determine the network topology in advance, which can be automatically obtained when the training process ends. As the simplified form of traditional SVM, the LSSVM encompasses advantages similar to those of SVM; in addition, it has the advantage that it is required to solve out a set of only linear equations (linear programming), which is much easier and simpler computationally.

Experimental procedures

Setups

The AE signal acquisition system used in the process of fiber tensile fracture in this study is shown in Figure 1. Equipped with a pneumatic stretching device (as shown in Figure 1(b)) to reduce the vibration of testing system, a low-noise AE signal acquisition platform (as shown in Figure 1(a)) was used in the experiments. Compressed by the air pump (as shown in Figure 1(c)), the air passes through the oil–water filter; the reversing valve is responsible for controlling the expansion and contraction of the cylinder which drives the connecting rod to grip the fiber for the stretching. Passing through the air compressor, the air is pumped into the oil–water filter for the purpose of purification and separation. In order to obtain both constant speed and pressure for the air flow, the pressure control valve (IR2000-02, SMC Ltd., China) was employed. The fiber chuck is connected to the cylinder piston rod through a nylon rod, and the fiber stretching can be realized by means of the contraction of cylinder piston rod. The polyvinylidene fluoride (PVDF) sensor (Jinzhou Kexin Electronic Materials Co., Ltd., China) is fixed between two fiber chucks, collecting the AE signal during the fiber tensile fracture process. Furthermore, the aerodynamic technique is used to reduce the noise in AE experiments. The oil–water filter was employed to store the air, and the required experiments were carried out under the condition of power failure at the end of the air pumping so as to avoid electromagnetic interference generated by the power supply. With the cylinder as its stretching unit, the uniform testing speed of the system can be ensured by setting the air pressure. In addition, stable airflow velocity facilitates reduction of the vibration generated during the test and avoids additive noise. A spectrum data acquisition card (M2i.4911-Exp SN9200) with memory capacity of 1 gigabyte, used for the signal storage, was utilized. The parameters of AE signal acquisition instrument are shown in Table 1.

Figure 1.

AE signal acquisition system for fiber tensile fracture: (a) schematic diagram of testing system, (b) pneumatic stretching device, (c) air supply device.

Table 1.

Parameters of AE signal acquisition instrument

Product model	Analog signal input	Resolution	Sampling frequency
M2i.4911-Exp	4	16 (bit)	10 MHz

Specimens

Polyester fibers (1.33 dtex × 38 mm) and cotton fibers (1.21 dtex × 40 mm) were chosen for the experiments with the target of the acquisition of AE signal generated during the single-fiber tensile fracture process. Initially, a single polyester fiber was placed onto the glue-coated paper sheet, followed by natural stretching with a clamping distance of 2 cm; subsequently, the paper sheet was bonded (the sample length is 4 cm, and the width is 1 cm); after that, the fiber was placed between two chucks of the sensor and then subjected to tensile fracture treatment; the AE signal generated during the fiber tensile fracture process was automatically collected by the computer. The air pressure value was set to 0.04 MPa. Sixty single-fiber tensile fracture AE signals were collected for both polyester and cotton fibers. The clamping status of a single fiber is shown in Figure 2. All experimental materials were placed under standard experimental conditions (temperature of ±2℃, relative humidity of 65 ± 2%) for 24 h during the experiment; in this instance, the standard moisture regain can be reached.

Figure 2.

Schematic diagram of fiber clamping status.

Results and discussion

Noise control

AE, in essence, is a transient elastic wave phenomenon caused by the rapid release of energy. It is commonly recognized that most of the signals obtained by AE sensors are generated by noise. Under the action of amplifier, the noise levels of signal were increased during the A/D transformation and then stored in the computer. Due to low signal intensities, external factors would interfere with the weak AE signals generated during the fiber tensile fracture process, thus resulting in the annihilation of real signals. Moreover, the process of fiber tensile fracture is, in nature, a dynamic detection process which often happens in a split second; it is therefore important to ensure the real-time performance and integrity of signal acquisition. In view of this, the unsatisfactory SNR of the AE signal generated by the fiber tensile fracture should be improved to some extent.

As one of the most critical indicators for the evaluation of single-fiber performance, the strength was inspected by a single-fiber strength tester in this experiment. In the early stage of the test, the AE signal generated by the fiber tensile fracture was mainly collected by means of the application of an AE sensor to the single-fiber strength tester. The AE signal acquisition system based on an electronic single-fiber strength tester (XS (08) X, Shanghai Xusai Instrument Co., Ltd., China) is shown in Figure 3. A PVDF piezoelectric sensor is placed between two chucks of the single-fiber strength tester which is fixed to lateral wall of the tester by a trestle, as shown in Figure 3.

Figure 3.

AE signal acquisition system for fiber tensile fracture, based on an electronic single-fiber strength tester.

Influenced by the internal motor, the SNR will be reduced to a considerable degree during the detection of the AE signal generated during the process of fiber tensile fracture performed by the electronic single-fiber strength tester. Therefore, a low-noise AE signal detection system was proposed in this paper. The same single-fiber AE signals collected by the two testing systems were analyzed; in addition the noise was intercepted with the noise levels of the two testing systems, and was analyzed and presented in the form of marginal spectrums. AE signals generated by the polyester fiber tensile fracture were collected from different testing systems, as shown in Figure 4. Figure 4(a) shows the AE signal generated during the process of the single polyester fiber tensile fracture obtained by the pneumatic device, and Figure 4(b) displays AE signal generated during the process of the single polyester fiber tensile fracture collected from the electronic single-fiber strength tester. In addition, the MATLAB software (version No.: 2017b, USA) was used for undertaking processing including reading the signal data and drawing the graphics for these acquired AE signals.

Figure 4.

AE signals generated by single-fiber tensile fracture and collected by different testing systems: (a) pneumatic device; (b) electronic single-fiber strength tester.

The AE signal obtained during the fiber tensile fracture process at the center of 0.2 s was set as the center point, and then 100,000 points from both forward and backward directions respectively (200,000 points in total) were intercepted. The intercepted noise signals are shown in Figure 5. Taking the noise signal as the source signal, the amplitude of each frequency in the noise signal was analyzed and presented in the form of marginal spectrum and the noise levels of the two acquisition systems were compared.

Figure 5.

Noise signals intercepted from AE signals generated by single-fiber tensile fracture :(a) pneumatic device; (b) electronic single-fiber strength tester.

Because AC power is required by the electronic single-fiber strength tester, the power frequency noise generated by it would radiate outward, thus affecting the noise level during the signal acquisition process. Spectrum analysis of the intercepted signals collected by both acquisition platforms (i.e. pneumatic device and electronic single-fiber strength tester) was carried out, as shown in Figure 6. Power frequency noise is mainly distributed at 50 Hz; the noise energy is mainly concentrated in the low-frequency part and distributed within the range of 0–100 Hz frequency band. Figure 6 reflects the signal energy distributions of the two devices within the range of 0–1000 Hz, with both peaks at 50 Hz. What can be clearly observed in the figure is that peak energy of 37.5 mv can be obtained from the spectrum of AE signal generated by traditional device, while peak energy of 25 mv can be viewed from that of pneumatic device. It can be seen that the noise level of pneumatic AE signal acquisition device under the condition of power failure is significantly reduced, which proves its practicability in noise reduction.

Figure 6.

Low-frequency bands of Hilbert marginal spectrums of noise signals acquired from different platforms.

Figure 7 shows the high-frequency bands of Hilbert marginal spectrums of noise signals acquired from the different platforms. By performing spectrum analysis on the intercepted noise portion (Hilbert marginal spectrum), it can be seen that the level of the noise gathered by the electronic single-fiber strength tester is almost entirely higher than that collected by the pneumatic device, with a sharp peak between 1 MHz and 3 MHz, especially at 2.6 MHz. The level of the noise obtained by the electronic single-fiber strength tester is nearly five times of that of the pneumatic device; the reason for this phenomenon is that the mechanical vibration generated by the servo motor of the electronic tester is larger than that from the pneumatic device. The dedicatedly designed pneumatic device presents small vibration amplitude and low noise level, contributing to the feature extraction and pattern recognition of the subsequent AE signal produced during the fiber tensile fracture process.

Figure 7.

High-frequency bands of Hilbert marginal spectrums of noise signals acquired from different platforms.

Based on the analysis of Figures 6 and 7, it can be seen that the single-fiber signal noises collected by the PVDF sensor are mainly concentrated in the low-frequency range, with 100 times of noise level difference. The power frequency noise, which was generated by the AC power and electrical motor, was deemed as the noise source. With regard to the PVDF sensor, it was mainly used for the acquisition of the signal with strong SNR. Owing to the low effective signal energy amplitude and rather high noise level, it is prone to annihilate the signal during the process of weak signal acquisition (such as signal acquisition during single-fiber tensile fracture). As per the characteristics of the PVDF sensor, it is recommended that the AC power and electrical motor be avoided in pneumatic stretching device, so as to reduce the power frequency noise and improve the SNR.

Feature extraction

Making use of the constructed EEMD–PCA–LSSVM model, 60 signals were collected from each polyester fiber and fine lint cotton fiber and a total of 10,000 points were intercepted from the starting point of fiber tensile fracture. The intercepted portions from the signals generated during the process of tensile fracture of both fibers are shown in the red frames in Figure 8. Intended for constructing time spectrums and extracting characteristic parameters from intercepted signals, both the EEMD processing and HHT processing were carried out on these signals. Figure 8(a) shows the AE signal generated during the process of polyester fiber tensile fracture and Figure 8(a’) displays the intercepted part of this signal. Figure 8(b) presents the AE signal generated during the process of cotton fiber tensile fracture and Figure 8(b’) reveals the intercepted part of this signal. In the process of fiber tensile fracture, signal generation starts from the breaking of macromolecular chains inside the fiber. As can be seen from Figure 8(a’) and Figure 8(b’), significant signal fluctuations appear when the tensile fracture of both fibers occurs. However, due to the different molecular structures of the different fibers, the AE signals generated by their fracture will exhibit different frequencies and energies. The initial part of fiber tensile fracture in the whole signal was taken as an effective signal to distinguish different fibers, and the characteristic parameters were extracted by the EEMD–HHT method.

Figure 8.

AE signals generated by fiber tensile fracture: (a) polyester fiber, (a’) intercepted part from (a), (b) cotton fiber, (b’) intercepted part from (b).

The decomposition steps of EEMD are as follows:

Add white noise sequence to target data;

Decompose the sequence which is added with white noise into IMF;

Repeat Step 1 and Step 2 when adding a different white noise sequence;

Take the mean value of each IMF as the final result.

The HHT is applied to each sequence of IMFs obtained from the EEMD, and each instantaneous frequency of IMFs is calculated. Then the signal can be expressed as:

x (t) = R_{e} \sum_{i = 1}^{n} a_{i} (t) e^{j \int w_{i} (t) dt}

(1)

In Equation (1), where R_e represents the real part, rn is omitted from the equation because it is a monotone function or a constant. Equation (1) can be used to express signal amplitude as a function of time–instantaneous frequency in three-dimensional space; moreover, the signal amplitude can also be expressed as a contour line on time–frequency surface. After carrying out these treatments, the amplitude distribution in the time–frequency space can be defined as the Hilbert time spectrum $H (ω, t)$ .

H (ω, t) = R_{e} \sum_{i = 1}^{n} a_{i} (t) e^{j \int w_{i} (t) dt}

(2)

The Hilbert spectrum of the AE signal generated by fiber tensile fracture contains three types of information: time series, frequency and energy. The time feature is divided into eight time periods in the time domain by HHT, which is distributed as T1–T8 according to the time sequence; the frequency feature is divided into 400 grid center frequencies; the energy characteristic refers to the amplitude in different time series and at different frequency distributions. Figure 9 shows the Hilbert spectrums of AE signals gathered from the tensile fracture of polyester fiber and cotton fiber. After processing by EEMD–HHT, the time–frequency characteristics of AE signals generated by the fiber tensile fracture were constructed. The Hilbert spectrum includes the time, frequency and energy information of the fiber fracture signal. Regarding the Hilbert spectrum, the abscissa represents the time, the ordinate represents the frequency, the color indicates the energy intensity (namely, the brighter the color is, the higher the energy intensity is). As shown in Figure 9, the fiber fracture is accompanied by energy release and it shows a high concentration of energy in the Hilbert spectrum.

Figure 9.

Hilbert spectrums of AE signals gathered from fiber tensile fracture: (a) polyester fiber tensile fracture; (b) cotton fiber tensile fracture.

As can be seen from the Hilbert spectrum, the energy (amplitude) increases gradually with time series in the whole frequency band, which directly represents the process of gradual destruction of the fiber under the action of external forces. With darker color comes larger signal amplitude, indicating the greater energy which is released when the fiber breaks. From the perspective of the time axis, different fibers release the energies at different time series. The energy release during polyester fiber fracture is mainly concentrated in the T5–T8 time series, while the energy release during cotton fiber fracture is mainly concentrated in the T4–T6 time series. A conclusion also can be drawn that the energy release during polyester fiber fracture is significantly higher than that of cotton fiber.

However, it is commonly known that the large feature dimension is not conducive to the recognition executed by the classifier. Aimed at dimension reduction from high to low, PCA can be performed to select the characteristic parameters with large contribution rate in the time–frequency characteristics of fiber signals. By taking the main characteristic parameters as the input data of the classifier, the classification effect of the classifier can be improved to a large extent. PCA, used as a statistical method, is realized by means of dimension reduction.¹⁹ By using an orthogonal transformation, the original random vector relevant to the component is transformed into a new random vector unrelated to the component, followed by the implementation of dimension reduction to multi-dimensional variable system so as to achieve the goal of transforming the variable system with high dimensionality into a new one with low dimensionality. Its representation in algebra refers to the transformation of the covariance matrix of the original random vector into a diagonal matrix.

The time–frequency–energy information can be obtained from the Hilbert spectrum of the AE signal generated by single-fiber fracture. For example, the three-dimensional matrix data of polyester fiber is shown in Table 2. The signal energy data in different time series and at different frequency distributions is with extremely high dimensionality, which is not conducive to the subsequent analysis; in view of this, the dimension reduction is required by the energy information. Thus, a 400 × 8 data matrix, containing energy information of the signal, is constructed. The results of characteristic parameters of polyester fiber and cotton fiber after PCA treatment are shown in Table 3 and Table 4, respectively, and it can be found that the cumulative contribution rate of the first three principal components after PCA dimension reduction treatment reaches over 85%. The first three principal components are extracted as a new feature set, which reduces the interference of other components and highlights useful information features. Taking the data matrix with the first three contribution rates as the input matrix of the separator, the computation amount of the classifier can be effectively reduced.

Table 2.

A three-dimensional eigenmatrix of polyester fiber

Serial number	Grid center frequency (MHz)	Time serial number
Serial number	Grid center frequency (MHz)	T1	T2	T3		T7	T8
1	0.000625	176742.3	12641.0	17804.6	…	1370.9	476.1
2	0.001875	5868.4	5258.7	26793.7	…	2508.0	995.6
3	0.003125	1348.1	1576.2	9347.4	…	2339.1	943.0
4	0.004375	757.8	956.0	10064.7	…	2139.8	711.7
…	…	…	…	…	…	…	…
400	0.499375	175195.6	14193.4	20362.8	…	1505.2	584.9

Table 3.

Contribution rates of principle components for polyester fiber

No.	Eigenvalue	Individual contribution rate (%)	Accumulative contribution rate (%)
1	128.614	58.371	58.371
2	58.762	18.273	76.644
3	43.266	9.087	85.731

Table 4.

Contribution rates of principle components for cotton fiber

No.	Eigenvalue	Individual contribution rate (%)	Accumulative contribution rate (%)
1	99.94	61.237	61.237
2	34.844	20.446	81.683
3	28.024	11.527	93.21

Identification via classification

SVM is a supervised learning method which has been widely used in statistical classification and regression analysis. It improves the generalization ability of the machine by seeking to minimize the structural risk and then minimize the empirical risk and confidence range, achieving the goal of obtaining good statistical laws in the case of a small number of statistical samples. Generally speaking, it is a binary classification model whose basic model is defined as the linear classifier with the largest spacing in the feature space, which can be converted into solving a quadratic programming problem. The LSSVM uses the least square linear system as the loss function, taking the place of traditional quadratic programming method due to its excellent superiority. It is required only to directly solve the problem of linear equations instead of quadratic programming, avoiding the disadvantages of SVM and speeding up the operation.²⁰

As an important component of SVM, researchers have paid considerable attention to the kernel function. Since different kernel functions have their own advantages and disadvantages, the selection of the kernel function determines the performance of SVM during the construction of SVM. RBF kernel function is a kind of local kernel function with high local learning ability and fewer parameters to be determined during practical application.

The time–frequency characteristics of the two fibers after PCA dimension reduction were normalized, the samples

n_{1 - 1} \sim n_{1 - 40}

and

n_{2 - 1} \sim n_{2 - 40}

were used as training samples of both fibers, and the samples

n_{1 - 41} \sim n_{1 - 60}

and

n_{2 - 41} \sim n_{2 - 60}

were used as testing samples of both fibers, respectively. SVM was established in this study, with the objective of the classification and identification of the fractures of different fibers. Based on fiber types, the LSSVM recognition machine consisting of two LSSVM classifiers was constructed. The radial basis kernel function was selected as the kernel function, and the penalty factor and kernel parameters were set to 2. The test results are shown in Table 5, and the recognition rate is shown in Table 6.

Table 5.

Recognition results of acoustic emission signals during single-fiber tensile fracture

Testing system			Forecast category
Testing system			Polyester fiber	Cotton fiber
Pneumatic device	Actual category	Polyester fiber	58	4
Pneumatic device		Cotton fiber	2	56
Electronic single-fiber strength tester		Polyester fiber	42	13
Electronic single-fiber strength tester		Cotton fiber	18	47

Table 6.

Acoustic emission signal recognition rate during single-fiber tensile fracture

Testing system	Fiber	Data set	Accuracy (%)	Overall accuracy (%)
Pneumatic device	Polyester fiber	60	96.67	95
Pneumatic device	Cotton fiber	60	93.33	95
Electronic single-fiber strength tester	Polyester fiber	60	70	74.17
Electronic single-fiber strength tester	Cotton fiber	60	78.33	74.17

It can be seen from Table 6 that the specially designed model possesses preferable recognition rates for AE signals generated by tensile fracture of both polyester fiber and fine lint cotton fiber, with an overall recognition rate of 95%. It is good that most of the testing results are consistent with the actual results. It is found that the LSSVM classifier is quite qualified for realizing the classification and recognition of AE signals generated by the tensile fracture of different fibers, with satisfactory results. Comparing the electronic single-fiber strength tester with the pneumatic device, the overall identification rate is improved from 74.17% to 95%. It can be concluded that the pneumatic device possesses the capacity of lowering the interference from noise, making the collected signals purer and improving the recognition rates of different kinds of fibers.

Conclusions

In view of the shortcomings of testing technology used during the single-fiber tensile fracture process, the AE technology was recommended to analyze the fracture of different single fibers. The idea is that different types of fiber have different internal molecular structures, and that the AE signal generated during fiber tensile fracture contains different characteristics. The traditional methods used for AE signal analysis focus on only parameter analysis, and exclude analysis on essential characteristics of the AE signal. In this paper, the time–frequency decomposition method (namely, HHT), which has been extensively used in current analysis on non-linear and non-stationary signal, was utilized to decompose the signal generated during single-fiber tensile fracture. Moreover, the PCA was employed to reduce the dimension of time–frequency information of the signal. The extracted single-fiber characteristic information was trained by the LSSVM. Therefore, the AE signal analysis model for single-fiber tensile fracture based on HHT–PCA–LSSVM was built, and the pattern recognitions of single-fiber tensile fracture of different materials were realized. In addition, this paper took the influence of noise on AE signal acquisition into full consideration and dedicatedly designed a low-noise AE signal acquisition device for single-fiber tensile fracture that can effectively reduce the influence of electromagnetic noise and mechanical vibration. The results showed that the established AE signal analysis model for single-fiber tensile fracture can better characterize different kinds of fiber fracture, and the recognition rate of single-fiber tensile fracture signal increased from 74% to 95% after reducing the influence of noise. This study provides technical support for research in fiber and yarn technology, and it also characterizes the fiber tensile fracture, which is difficult to observe by AE technology.

Footnotes

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by SUES Industrial Supporting Project (18FZ-008), Shanghai Local Capacity-Building Project (No. 19030501200) and Shanghai University of Engineering ScienceTalents Zhihong Project (2017RC432017).

ORCID iDs

Binjie Xin

Yuansheng Zheng

References

Furch

Glos

. Identifying technical condition of vehicle gearbox using acoustic emission. Materials Science & Engineering Conference Series 2017; 212: 012001.

Zhou L, Duan F, Mba D, et al. A comparative study of helicopter planetary bearing diagnosis with vibration and acoustic emission data. In: IEEE International Conference on Prognostics & Health Management. IEEE (ICPHM), 01 June 2017, DOI: 10.1109/ICPHM.2017.7998336.

Mahmoud

Vlasic

Mazal

, et al. Leakage analysis of pneumatic cylinders using acoustic emission. Insight - Non-Destruct Test Condition Monitoring J 2017; 59(9): 500–505.

Zhao J, Sheng H and Zhou X. Study on the application of acoustic emission testing technique in monitoring 16Mn steel welding defects. In: International Conference on Advanced Mechatronic Systems, 01 Nov 2017. DOI: 10.1109/ICAMechS.2016.7813467.

Tang

Wang

Lei

, et al. Evolution characteristics of reservoir rock cracks based on acoustic emission fractal parameter analysis. Geophys Prospect Petroleum J. Epub ahead of print June 2017.

Mothe GS, Dandawate YH and Achwal V. Signal processing system design for acoustic emission signatures detetction and analysis. In: 2017 2nd IEEE International Conference on Recent Trends in Electronics, Information & Communication Technology (RTEICT), 01 May 2017. DOI: 10.1109/RTEICT.2017.8256823.

Danyuk

Rastegaev

Pomponi

, et al. Improving of acoustic emission signal detection for fatigue fracture monitoring. Proc Eng J 2017; 176: 284–290.

Angelopoulos

Papaelias

. Automated analysis of acoustic emission signals based on the ISODATA algorithm. Insight - Non-Destruct Test Condition Monitoring J 2018; 60: 3–3.

Rodríguez

Celestino

. Application of acoustic emission monitoring and signal analysis to the qualitative and quantitative characterization of the fracturing process in rocks. Eng Fracture Mechanics J 2018. S0013794418302157.

10.

Gong

Song

, et al. Precursory waves and eigenfrequencies identified from acoustic emission data based on singular spectrum analysis and laboratory rock-burst experiments. Int J Rock Mechan Mining Sci 2017; 91: 155–169.

11.

Zhou

Zhao

Zhang

, et al. Cluster analysis of acoustic emission signals and deformation measurement for delaminated glass fiber epoxy composites. Composite Struct J. 2018; 195: 349–358.

12.

Malgorzata

Wozniak

Ranachowski

, et al. Analysis of acoustic emission signals at austempering of steels using neural networks. Metal Mater Int J 2017; 23(3): 1–8.

13.

Huang

Shen

Long

, et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis. Proc A J 1998; 454(1971): 903–995.

14.

Pai

Palazotto

. HHT-based nonlinear signal processing method for parametric and non-parametric dentification of dynamical systems. Int J Mechan Sci 2008; 50(12): 1619–1635.

15.

Daetig

Schlurmann

. Performance and limitations of the Hilbert–Huang transformation (HHT) with an application to irregular water waves. Ocean Eng J 2004; 31(14/15): 1783–1834.

16.

Zafar

Kamal

Sheikh

, et al. A neural network-based approach for background noise reduction in airborne acoustic emission of a machining process. J Mechan Sci Technol 2017; 31(7): 3171–3182.

17.

Guo

Wen

, et al. Adaptive noise cancellation based on EMD in water-supply pipeline leak detection. Measurement J 2016; 79: 188–197.

18.

Jena

Sahoo

Panigrahi

. Gear fault diagnosis using active noise cancellation and adaptive wavelet transform. Measurement J 2014; 47: 356–372.

19.

Yang

Dong

Lim

LHI

, et al. Analyzing big time series data in solar engineering using features and PCA. Solar Energy J 2017; 153: 317–328.

20.

Zhang

Tian

Song

, et al. Research on an optical e-nose denoising method based on LSSVM. Optik J 2018; 168: 118–126.