ACOUSTIC EMISSION CHARACTERISTICS OF USED 70 MPA TYPE

IV HYDROGEN STORAGE TANKS DURING HYDROSTATIC BURST

TESTS

Dongliang Wang1, Binbin Liao1, Chunyong Hao1, Ange Wen1, Jinyang Zheng1, *, Peng Jiang1, 2, *,

Chaohua Gu1, Ping Xu3, Qianghua Huang4 1 Institute of Process Equipment, Zhejiang University, Hangzhou 310027, PR China

2 Northeast Petroleum University, Heilongjiang Daqing 163318, PR China 3 Institute of Applied Mechanics, Zhejiang University, Hangzhou 310027, PR China

4 China Special Equipment Inspection and Research Institute, Beijing 100029, PR China

*Corresponding author. E-mail: jyzh@zju.edu.cn (Jinyang Zheng), jpnepu@163.com (Peng Jiang)

Abstract

Currently, the periodic inspection of composite tanks is typically achieved via hydrostatic test combined

with internal and external visual inspections. Acoustic emission (AE) technology demonstrates a

promising nondestructive testing method for damage mode identification and damage assessment. This

study focuses on AE signals characteristics and evolution behaviors for used 70 MPa Type IV hydrogen

storage tanks during hydrostatic burst tests. AE-based tensile tests for epoxy resin specimen and carbon

fiber tow were implemented to obtain characteristics of matrix cracking and fiber breakage. Then, broad-

band AE sensors were used to capture AE signals during multi-step loading tests and hydrostatic burst

tests. K-means ++ algorithm and wavelet packet transform are performed to cluster AE signals and

verify the validity. Combining with tensile tests, three clusters are manifested via matrix cracking,

fiber/matrix debonding and fiber breakage according to amplitude, duration, counts and absolute energy.

The number of three clustering signals increases with the increase of pressure, showing accumulated

and aggravated damage. The sudden appearance of a large number of fiber breakage signals during

hydrostatic burst tests suggests that the composite tank structure is becoming mechanically unstable,

namely the impending burst failure of the tank.

1. INTRODUCTION

International community is facing serious global problems in the present age, such as global warming,

fossil fuel depletion and environmental pollution, which are directly associated with excessive usage of

fossil fuels. As a clean, efficient and renewable energy, hydrogen energy will play an important role in

the future sustainable energy system [1]. Hydrogen fuel cell vehicles are the main way of hydrogen

energy utilization and are moving from demonstration to commercialization. Composite high-pressure

tanks are the most widely used and mature as hydrogen storage containers in hydrogen fuel cell vehicles

[2,3], due to many advantages such as high strength, high stiffness-to-weight ratio and excellent fatigue

resistance [4,5].

Currently, the periodic inspection and testing of composite tanks is typically achieved via hydrostatic

test combined with internal and external visual inspections, according to the standard ISO 11623 [6] and

CGA C-6.4 [7]. The damage levels obtained from internal and external visual inspections determine the

service or rejection of composite tanks. However, this method depends a lot on the artificial experience.

It is necessary to develop alternatives to visual inspection for hydrogen storage tanks.

Acoustic emission (AE) technology has shown great capability in damage detection of composite

materials in the last decades [8-11]. For example, Sayar et al. [11] analyzed the clustering effect on

damage mechanisms under tensile loading in open-hole laminated carbon/epoxy composites using AE

method. They obtained the frequency ranges for matrix cracking, fiber breakage, fiber pull-out,

fiber/matrix debonding and delamination. Roundi et al. [12] studied damage progression in glass/epoxy

composites during static and fatigue tensile tests using AE method. They found that the matrix micro-

cracks were the most dominant damage mechanisms during the static and cyclic tests, and few signals

were detected representing fiber breakage at the end of the tests. Sause et al. [13] proposed an AE-based

approach to predict failure load for fiber-reinforced composites using the Felicity ratio (FR) combined

with an artificial neural network and a simple linear extrapolation.

Due to great capability in composite materials damage detection of AE technology, researchers began

to utilize AE technology to monitor damage for composite tanks. Chou et al. [14] explored AE-based

damage detection in carbon fiber composite pressure vessels under constant and cyclic pressure

combined with laminate tests. Their results indicated that a sudden rise in the number of AE hits was a

more reliable indicator of the mechanical integrity of composite pressure vessels comparing to the total

number of AE hits. Lin et al. [15] studied thermo-mechanical properties of composite pressure vessel

under hydraulic and atmospheric fatigue cycling using AE technology. Their results indicated that there

were more AE events in the atmospheric fatigue test. Dahmene et al. [16] investigated the damage

evolution of composite pressure vessels during hydrogen filling via AE method. They found that Felicity

ratio could be regarded as a useful pass/fail criterion, but alternatives should be found due to the

complexity of the test procedure.

From above researches, AE technology shows a promising nondestructive testing method to define the

tank’s suitability for service. However, in order to protect AE sensors, these studies did not capture AE

signals during the composite tank burst process. AE signals during the composite tank burst process can

reflect damage characteristics and evolution behaviors, especially AE signals related to damage status

at the moment of the tank burst failure, which is important for damage progression assessment using AE

technology.

The purpose of this paper is to study AE signals characteristics and evolution behaviors in used 70 MPa

Type IV hydrogen storage tanks during hydrostatic burst tests. AE signals were detected on two tanks

during multi-step loading tests and hydrostatic burst tests using multi-step loading method. By

combining with AE-based tensile tests for epoxy resin specimen and carbon fiber tow, AE signals are

associated with damage mechanisms (i.e. matrix cracking, fiber/matrix debonding and fiber breakage)

according to amplitude, duration, counts and absolute energy. Then, three clusters signals of two tanks

are performed using K-means++ algorithm. Wavelet packet transform is used to verify the peak

frequency ranges of three clusters. Finally, AE signals evolution behaviors with the change of pressure

are discussed. The characteristics of AE signals at the moment of the tank burst failure are also studied.

2. EXPERIMENTAL PROCEDURES

2.1 Hydrogen storage tanks and materials description

The test cylinders are two identical Type IV tanks having a capacity of 42 L, length of 900 mm, external

diameter of 310 mm, and composite wall thickness of 26 mm. The nominal working pressure is 70 MPa.

Before performing experiments, two tanks experienced the same driving distance of 46500 kilometers

installed on the hydrogen fuel cell vehicle, and then were subjected to hydraulic fatigue tests for 7500

cycles at 50 °C and 65 °C, respectively. Epoxy resin (Toray Industries Inc.) specimen tensile test was

conducted to obtain matrix cracking signals. The fiber tow used in tensile test was T700 carbon fiber

tow (Toray Industries Inc.). Each carbon fiber tow contains 100 carbon fiber monofilaments.

2.2 Experimental setups

In order to relate AE signals with damage mechanisms of matrix cracking and fiber breakage,

respectively, tensile tests for epoxy resin specimen and carbon fiber tow were performed. Epoxy resin

specimen tensile test was performed under the elongation rate of 1 mm/min using a SANS universal

testing machine. Carbon fiber tow tensile test was conducted using the MTI SEMtester tensile testing

bench with the maximum loading 4500 N. The elongation rate was 5 mm/min. According to obtained

AE signals characteristics, damage mechanisms in hydrogen storage tanks during multi-step loading

tests and hydrostatic burst tests can be identified and distinguished in the following study.

The multi-step loading tests and hydrostatic burst tests of two tanks were performed using a

multifunctional pressure testing system with the maximum pressure 200 MPa. Each tank was

pressurized in two stages using a multi-step loading approach. The multi-step loading approach consists

of four phases, namely pressure increasing, pressure holding, pressure unloading, and pressure holding,

which aims to obtain useful AE signals and eliminate water-filling noise interference. Tests arrangement

is listed in Table 1. First, the tank was pressurized from 0 to 158 MPa/150 MPa in the way of multi-step

loading, as shown in Fig.1a. Then, the tank was loaded from 0 to burst, as shown in Fig.1b.

Table 1. Hydrogen storage tanks tests arrangement

Tank Driving distance

(km)

Initial fatigue

cycles Test

Pressure range

(MPa) Test label

T1 46500 7500 (50 °C) Multi-step loading test 0-158 T1-M

Hydrostatic burst test 0 to burst T1-B

T2 46500 7500 (65 °C) Multi-step loading test 0-150 T2-M

Hydrostatic burst test 0 to burst T2-B

(a) (b)

Figure 1. Pressure loading approaches: (a) Multi-step loading tests, (b) Hydrostatic burst tests.

2.3 Acoustic emission monitoring

An AE monitoring system (Physical Acoustic Corporation) was used to monitor damage behaviors,

which includes an AEWin software, an eight-channel data acquisition module, transducers and

preamplifiers. For carbon fiber tow tensile test, the ends of the fiber tow were bonded with aluminum

sheets using the adhesive, and two Nano30 piezoelectric sensors were placed on the aluminum sheets,

as shown in Fig.2a. For resin tensile test, two WD piezoelectric sensors were directly placed on the

epoxy resin specimen, as shown in Fig.2b.

Figure 2. (a) Carbon fiber tow tensile test, (b) Epoxy resin specimen tensile test.

In multi-step loading tests, eight WD piezoelectric sensors with a frequency range of 100-1000 kHz

were used to capture AE signals. As shown in Fig.3a, six AE sensors were placed on the cylindrical part

of the tank, two AE sensors were placed on the dome area. One AE sensor was placed on the cylindrical

part of the tank to capture AE signals during hydrostatic burst tests, as shown in Fig.3b. The silicone

grease was used for the contact of sensor to the tank surface for good acoustic coupling. AE signals were

amplified 40 dB by preamplifiers. The sampling frequency is 1 MHz. Also, a threshold value of 40 dB

was used to filter background noise. Then, the pencil-lead breakage method was utilized for calibrating

AE sensors. To obtain pressure values and AE signals simultaneously, pressure transducers were used

to connect the AE monitoring system and the multifunctional pressure testing system.

Figure 3. Layout of AE sensors on the tank: (a) Multi-step loading tests, (b) Hydrostatic burst tests.

3. ANALYSIS METHOD

3.1 AE cluster analysis

During AE tests, many AE parameters are recorded in software, such as the rise time, the counts, the

energy, the duration, the amplitude, the peak frequency et al. To improve clustering quality and reduce

computational time, it is necessary to perform feature selection before cluster analysis. Principal

component analysis (PCA) is performed to select important feature parameters, combined with the

evaluation index of correlation coefficient [17]. As a multivariate analysis method, PCA transforms

multidimensional AE data into lower dimensions through an orthogonal linear transformation so that the

data can be better analyzed and visualized [17,18]. Correlation coefficient ranges between 0 to 1, which

represents the dependency of features on each other [19]. Feature selection is based on low dependency

on other features.

The K-means++ algorithm is a modified method of choosing centers for the k-means algorithm, which

is a centroid based and an iterative algorithm [17-21]. The k-means++ algorithm is described as follows:

(1) Randomly choose a center from among the data points;

(2) For each AE data point ( 1,2, , )ix i n= , compute the shortest distance ( )iD x from ix to the

closest center ( 1,2, )jB j k= we have already chosen;

(3) Use weighted probability distribution 2

2

1

( )

( )

i

n

i

i

D x

D x=

to choose one new data point ( 1,2, , )ix i n= at

random as a new center. Thus, weighted probability distribution is proportional to ( )iD x ;

(4) Repeat step 2 and 3 until k centers have been chosen;

(5) For each AE data point ( 1,2, , )ix i n= , compute the distance between ix and each center and

assign ix to the nearest cluster center;

(6) new jB locations are recalculated for the new clusters;

(7) Repeat step 5 and 6 until all cluster centers do not change anymore.

Davies-Bouldin index is one of the clustering evaluation indexes. Davies-Bouldin index is based on a

ratio of within-cluster and between-cluster distances, which relates to cluster centroids [19]. The high-

quality cluster has the lower Davies-Bouldin index.

3.2 Wavelet packet transform

Wavelet packet transform (WPT) is a competent tool for AE data processing. In this study, WPT is used

to verify the validity of the cluster analysis. The wavelet has a limited frequency period with the mean

value of zero. If the energy function is limited and the functional integral is zero, the function of ( )t

is called the mother wavelet, as follows [22,23],

2( )t dt

+

−

(1)

( ) 0t dt+

−

= (2)

In Equation (1) and Equation (2), t is the time.

WPT decomposes a signal into detailed components (high-frequency components) and approximated

components (low-frequency components), which is beneficial to time-frequency localization analysis of

signals containing a large amount of medium and high frequency information. The frequency ranges of

the approximated and detailed components in each level can be expressed as [24]

10, 2

2

i

sf−

(3)

( 1)1 12 , 2

2 2

i i

s sf f− − −

(4)

In Equation (3) and Equation (4), sf is the sample rate, 2i is the number of components at level i.

Relative energy distribution at each level ( )j

iP t is exploited for characterization of failure modes,

which is expressed in Equations (5) to (7) as follows [24],

( )( ) 1, , 2

( )

j

j ii

i

Total

E tP t j

E t= = (5)

0

2( ) ( ( )) 1, , 2t

j j i

i i

t

E t f j

=

= = (6)

( ) ( ) 1, , 2j i

Total i

i

E t E t j= = (7)

In Equations (5) to (7), ( )f t is an AE signal, 1 j

i iE E express the energy of the components at level

i, TotalE is the total energy of AE signal, 1 j

i if f are the components of thi level of the decomposed

signal. More details about WPT method can be found in the literature [11,24].

4. RESULTS AND DISCUSSION

4.1 AE signals characteristics corresponding to specific damage mechanisms

To relate AE signals with specific damage mechanisms, such as matrix cracking and fiber breakage, AE

signals in each test of epoxy resin specimen and carbon fiber tow under tensile loading were recorded

and analyzed. The amplitude, counts, duration and absolute energy of AE signals were analyzed, as

shown in Fig. 4-5. Fig. 4 shows the AE signals characteristics for epoxy resin specimen tensile test.

According to Fig. 4, AE signals representing the matrix cracking have low amplitude (40-58 dB), short

duration (below 2500 us), few counts (below 60) and low absolute energy (40-2.0×103 aJ). The

localization of AE signals is consistent with the fracture location of the specimen, as shown in Fig. 4c.

(a) (b)

(c)

Figure 4. AE signals characteristics for epoxy resin specimen tensile test: (a) Correlated distribution of

counts and amplitude with duration, (b) Variation of counts and amplitude with testing time, (c)

Localization of AE signals.

Fig.5 shows the AE signals characteristics for carbon fiber tow tensile test. The amplitudes of obtained

AE signals change in a wide range of 40-99 dB. It is noteworthy that carbon fiber monofilaments were

in a state of relaxation, intertwining and even bend prior to tensile test. Besides, carbon fiber

monofilaments did not break simultaneously. Thus, a large number of low amplitude (40-77.5 dB) and

low absolute energy (<5.0×104 aJ) signals were captured due to fiber friction during the tensile test, as

shown in Fig. 5b and c. The AE signals with amplitude of 77.5-99 dB are related to fiber breakage,

showing longer duration and higher absolute energy (Fig. 5a and c). And Fig. 5c demonstrates that the

AE signals with amplitude of 77.5-99 dB are distributed at about 4.0 cm (abscissa values), which is

consistent with the fracture location of the fiber tow. Based on the test results, AE signals characteristics

corresponding to matrix cracking and fiber breakage can be summarized in Table 2, which are consistent

with the literatures [10,12,20,25,26].

(a) (b)

(c)

Figure 5. AE signals characteristics for carbon fiber tow tensile test: (a) Correlated distribution of

counts and amplitude with duration, (b) Variation of counts and amplitude with testing time, (c)

Localization of AE signals.

Table 2. AE signals characteristics corresponding to specific damage mechanisms

Damage mechanisms Amplitude (dB) Duration (us) Counts Absolute energy (aJ)

Matrix cracking 40-58 <2500 <60 40-2.0×103

Fiber breakage 77.5-99 850-8569 50-210 5.0×104-1.0×106

Fiber/matrix debonding [25] 50-80 - - 1.0×103-6.5×104

4.2 Damage evolution behaviors for hydrogen storage tanks

Four AE parameters such as amplitude, duration, counts and absolute energy, are chosen as feature

parameters by correlation coefficient analysis. According to obtained AE signals characteristics (Table

2) and our previous study [25], AE signals obtained from multi-step loading tests and hydrostatic burst

tests are clustered into three clusters using K-means++ algorithm, namely matrix cracking, fiber

breakage and fiber/matrix debonding. In order to verify the validity of the cluster analysis, wavelet

packet transform is used to obtain frequency ranges of three clusters. AE signals are decomposed to 3

different levels and 8 components. Each component of the 3rd level of three clusters is shown in Fig. 6.

Fig. 7 illustrates the wavelet component energy distribution. Fig. 6-7 show that the dominant energy of

matrix cracking, fiber/matrix debonding and fiber breakage is in component 1, component 2 and

component 4, respectively. Table 3 shows the frequency ranges for wavelet components. Thus, the peak

frequency ranges of matrix cracking, fiber/matrix debonding and fiber breakage are about 0-125 kHz,

125-250 kHz and 375-500 kHz, respectively. The corresponding relationships between frequencies and

damage mechanisms are consistent with reported literatures [18,25,27,28].

(a) (b)

(c)

Figure 6. The WPT decomposition (voltage [V] versus time [us]): (a) The matrix cracking signal, (b)

The fiber/matrix debonding signal, (c) The fiber breakage signal.

(a) (b)

(c)

Figure 7. The wavelet component energy distribution: (a) The matrix cracking signal, (b) The

fiber/matrix debonding signal, (c) The fiber breakage signal.

Table 3. Frequency ranges for wavelet components

Components 1 2 3 4 5 6 7 8

Frequency

(kHz) 0-125 125-250 250-375 375-500 500-625 625-750 750-875 875-1000

Fig. 8 shows the distribution of AE signals during multi-step loading tests. Due to the equipment

problems, the pressure values and AE signals were not obtained in Stage X, as shown in Fig. 8b. In Fig.

8, there are lots of matrix cracking signals during multi-step loading tests, while fiber breakage signals

are the least. There are almost no AE signals before 70 MPa, indicating that there is no damage in tanks.

Besides, fiber breakage signals in multi-step loading tests are insignificant and scattered, indicating that

the tank has good strength at this time.

(a) (b)

Figure 8. Variation of amplitude and pressure with testing time for (a) T1-M and (b) T2-M tests.

Fig. 9 illustrates the change in amplitude and pressure with time during hydrostatic burst tests. According

to Fig. 9a and b, the pressure-time curve of tank T1 has three pressure increasing phases (0-158 MPa,

158-168 MPa and 168-176.8 MPa) and two pressure holding phases (158 MPa and 168 MPa), the

pressure-time curve of tank T2 has four pressure increasing phases (0-150 MPa, 150-158 MPa, 158-168

MPa and 168-169.6 MPa) and two pressure holding phases (150 MPa, 158 MPa and 168 MPa). Due to

water-filling noise interference in the first pressure increasing phase (0-158 MPa for tank T1, 0-150 MPa

for tank T2), the AE signals at this phase were not analyzed. Fig. 9a and b show that the number of AE

signals is not large in total during hydrostatic burst tests. This can be explained by the quantity of AE

sensor (Fig. 3b) and the second loadings applied to the composite materials. The number of three

clustering signals increases with the increase of pressure, showing accumulated and aggravated damage.

Besides, the pressure distribution values for fiber/matrix debonding signals are generally greater than

those for matrix cracking. It can be found that there are no fiber breakage signals in tank T1 before 168

MPa, while this pressure is 161 MPa for tank T2, indicating that the fiber breakage is mainly caused by

high loading. Fiber breakage signals of tank T2 appear earlier than those of tank T1, which may be due

to higher temperature during fatigue test (Table 1).

(a) (b)

Figure 9. Variation of amplitude and pressure with testing time for (a) T1-B and (b) T2-B tests.

Fig. 10a and b show the change in events and pressure with time during hydrostatic burst tests, which

can be separated into three stages. The matrix cracking signals and fiber/matrix debonding signals

increase slowly at stage I. At stage II, matrix cracking signals increase quickly, fiber/matrix debonding

signals increase obviously, and fiber breakage signals begin to appear, but in small quantities. Finally,

matrix cracking signals, fiber/matrix debonding signals and fiber breakage signals increase sharply. It

can be concluded that the stage I, stage II and stage III are corresponding to damage initiation, damage

accumulation and burst failure, respectively. The burst pressure of tank T1 and tank T2 is 176.8 MPa

and 169.6 MPa, respectively. The pressure of tank T1 and tank T2 when reaching stage III is 168 MPa

and 161MPa, respectively, which both are about 95 percent of the burst pressure. Therefore, a large

number of three clustering signals appear simultaneously at the last moment, especially the rapid

increase of fiber breakage signals, indicating that the composite tank structure is becoming mechanically

unstable, namely the impending burst failure of the tank. Fig. 11 shows the appearance of hydrogen

storage tanks after burst failure. The tanks were exploded into several parts, and three clustering signals

can be clearly observed.

(a) (b)

Figure 10. Variation of events and pressure with testing time for (a) T1-B and (b) T2-B tests.

(a) (b)

Figure 11. The appearance of hydrogen storage tanks after burst failure: (a) tank T1, (b) tank T2.

5. CONCLUSIONS

In this study, AE signals characteristics and evolution behaviors in used 70 MPa Type IV hydrogen

storage tanks during hydrostatic burst tests are investigated. A multi-step loading method is applied to

capture AE signals of two tanks during multi-step loading tests and hydrostatic burst tests. The

characteristics of matrix cracking and fiber breakage were obtained from AE-based tensile tests for

epoxy resin specimen and carbon fiber tow. K-means ++ algorithm and wavelet packet transform are

performed to cluster AE signals and verify the validity. Finally, the characteristics and evolution

behaviors of AE signals are explored with the increasing pressure during multi-step loading tests and

hydrostatic burst tests. Obtained conclusions could be listed as follows:

(1) The signal characteristics (amplitude, duration, counts and absolute energy) of matrix cracking and

fiber breakage are obtained based on tensile tests for epoxy resin specimen and carbon fiber tow. The

amplitude, duration, counts and absolute energy of matrix cracking are generally the lowest, while those

corresponding to fiber breakage are the highest.

(2) The matrix cracking signals appear continuously and extensively during multi-step loading tests and

hydrostatic burst tests. The pressure distribution values for fiber/matrix debonding signals are generally

greater than those for matrix cracking. The number of fiber breakage signals are the least and appears at

the highest pressure.

(3) A large number of three clustering signals appear simultaneously at the last moment, especially the

rapid increase of fiber breakage signals, indicating that the composite tank structure is becoming

mechanically unstable, namely the impending burst failure of the tank.

ACKNOWLEDGEMENTS

All authors wish to express their sincere thanks to the support of the National Key Research and

Development Program of China (No. 2017YFC0805601).

REFERENCES

1. VEZIRO LU, T. N. and AHI N, S. 21st Century’s energy: Hydrogen energy system. Energy

Conversion and Management, 49, No. 7, 2008, pp. 1820-1831.

2. ZHENG, J., LIU, X., XU, P., LIU, P., ZHAO, Y. and YANG, J. Development of high pressure

gaseous hydrogen storage technologies. INTERNATIONAL JOURNAL OF HYDROGEN ENERGY,

37, No. 1, 2012, pp. 1048-1057.

3. DURBIN, D. J. and MALARDIER-JUGROOT, C. Review of hydrogen storage techniques

for on board vehicle applications. INTERNATIONAL JOURNAL OF HYDROGEN ENERGY, 38,

No. 34, 2013, pp. 14595-14617.

4. BARTHELEMY, H., WEBER, M. and BARBIER, F. Hydrogen storage: Recent improvements and

industrial perspectives. INTERNATIONAL JOURNAL OF HYDROGEN ENERGY, 42, No. 11,

2017, pp. 7254-7262.

5. TAKEICHI, N., SENOH, H., YOKOTA, T., TSURUTA, H., HAMADA, K., TAKESHITA, H. T.,

TANAKA, H., KIYOBAYASHI, T., TAKANO, T. and KURIYAMA, N. "Hybrid hydrogen storage

vessel", a novel high-pressure hydrogen storage vessel combined with hydrogen storage material.

INTERNATIONAL JOURNAL OF HYDROGEN ENERGY, 28, 2003, pp. 1121-1129.

6. Gas cylinders - Composite construction - Periodic inspection and testing., ISO 11623:2015.

7. METHODS FOR EXTERNAL VISUAL INSPECTION OF NATURAL GAS VEHICLE (NGV)

AND HYDROGEN GAS VEHICLE (HGV) FUEL CONTAINERS AND THEIR

INSTALLATIONS., CGA C-6.4-2012.

8. BRUNNER, A. J. Identification of damage mechanisms in fiber-reinforced polymer-matrix

composites with Acoustic Emission and the challenge of assessing structural integrity and service-

life. CONSTRUCTION AND BUILDING MATERIALS, 173, 2018, pp. 629-637.

9. BAKER, C., MORSCHER, G. N., PUJAR, V. V. and LEMANSKI, J. R. Transverse cracking in

carbon fiber reinforced polymer composites: Modal acoustic emission and peak frequency analysis.

COMPOSITES SCIENCE AND TECHNOLOGY, 116, 2015, pp. 26-32.

10. LIU, P. F., CHU, J. K., LIU, Y. L. and ZHENG, J. Y. A study on the failure mechanisms of carbon

fiber/epoxy composite laminates using acoustic emission. MATERIALS & DESIGN, 37, 2012, pp.

228-235.

11. SAYAR, H., AZADI, M., GHASEMI-GHALEBAHMAN, A. and JAFARI, S. M. Clustering effect

on damage mechanisms in open-hole laminated carbon/epoxy composite under constant tensile

loading rate, using acoustic emission. COMPOSITE STRUCTURES, 204, 2018, pp. 1-11.

12. ROUNDI, W., EL MAHI, A., EL GHARAD, A. and REBIERE, J. Acoustic emission monitoring

of damage progression in Glass/Epoxy composites during static and fatigue tensile tests. APPLIED

ACOUSTICS, 132, 2018, pp. 124-134.

13. SAUSE, M. G. R., SCHMITT, S. and KALAFAT, S. Failure load prediction for fiber-reinforced

composites based on acoustic emission. COMPOSITES SCIENCE AND TECHNOLOGY, 164,

2018, pp. 24-33.

14. CHOU, H. Y., MOURITZ, A. P., BANNISTER, M. K. and BUNSELL, A. R. Acoustic emission

analysis of composite pressure vessels under constant and cyclic pressure. Composites Part A:

Applied Science and Manufacturing, 70, No. Supplement C, 2015, pp. 111-120.

15. LIN, S., JIA, X., SUN, H., SUN, H., HUI, D. and YANG, X. Thermo-mechanical properties of

filament wound CFRP vessel under hydraulic and atmospheric fatigue cycling. COMPOSITES

PART B-ENGINEERING, 46, No. 3, 2013, pp. 227-233.

16. DAHMENE, F., YAACOUBI, S., EL MOUNTASSIR, M., LANGLOIS, C. and BARDOUX, O.

Towards efficient acoustic emission testing of COPV, without Felicity ratio criterion, during

hydrogen-filling. INTERNATIONAL JOURNAL OF HYDROGEN ENERGY, 41, No. 2, 2016, pp.

1359-1368.

17. XU, D., LIU, P. F., LI, J. G. and CHEN, Z. P. Damage mode identification of adhesive composite

joints under hygrothermal environment using acoustic emission and machine learning.

COMPOSITE STRUCTURES, 211, 2019, pp. 351-363.

18. LI, L., LOMOV, S. V., YAN, X. and CARVELLI, V. Cluster analysis of acoustic emission signals

for 2D and 3D woven glass/epoxy composites. COMPOSITE STRUCTURES, 116, 2014, pp. 286-

299.

19. CARVELLI, V., D'ETTORRE, A. and LOMOV, S. V. Acoustic emission and damage mode

correlation in textile reinforced PPS composites. COMPOSITE STRUCTURES, 163, 2017, pp. 399-

409.

20. OZ, F. E., ERSOY, N., MEHDIKHANI, M. and LOMOV, S. V. Multi-instrument in-situ damage

monitoring in quasi-isotropic CFRP laminates under tension. COMPOSITE STRUCTURES, 196,

2018, pp 163-180.

21. OZ, F. E., AHMADVASHAGHBASH, S. and ERSOY, N. Damage mode identification in

transverse crack tension specimens using acoustic emission and correlation with finite element

progressive damage model. Composites Part B: Engineering, 165, 2019, pp. 84-95.

22. AZADI, M., SAYAR, H., GHASEMI-GHALEBAHMAN, A. and JAFARI, S. M. Tensile loading

rate effect on mechanical properties and failure mechanisms in open-hole carbon fiber reinforced

polymer composites by acoustic emission approach. Composites Part B: Engineering, 158, 2019,

pp. 448-458.

23. RM, R. and AS, B. Wavelet transforms: introduction to theory and applications. Addison Wesley

Publishing Company, 1998, pp. 1-26.

24. FOTOUHI, M., SADEGHI, S., JALALVAND, M. and AHMADI, M. Analysis of the damage

mechanisms in mixed-mode delamination of laminated composites using acoustic emission data

clustering. JOURNAL OF THERMOPLASTIC COMPOSITE MATERIALS, 30, No. 3, 2017, pp.

318-340.

25. LIAO, B. B., WANG, D. L., HAMDI, M., ZHENG, J. Y., JIANG, P., GU, C. H. and HONG, W. R.

Acoustic emission-based damage characterization of 70 MPa type IV hydrogen composite pressure

vessels during hydraulic tests. INTERNATIONAL JOURNAL OF HYDROGEN ENERGY. 2019.

26. MAREC, A., THOMAS, J. H. and EL GUERJOUMA, R. Damage characterization of polymer-

based composite materials: Multivariable analysis and wavelet transform for clustering acoustic

emission data. MECHANICAL SYSTEMS AND SIGNAL PROCESSING, 22, No. 6, 2008, pp. 1441-

1464.

27. GUTKIN, R., GREEN, C. J., VANGRATTANACHAI, S., PINHO, S. T., ROBINSON, P. and

CURTIS, P. T. On acoustic emission for failure investigation in CFRP: Pattern recognition and peak

frequency analyses. MECHANICAL SYSTEMS AND SIGNAL PROCESSING, 25, No. 4, 2011, pp.

1393-1407.

28. DEGROOT, P. J., WIJNEN, P. and JANSSEN, R. Real-time frequency determination of acoustic

emission for different fracture mechanisms in carbon epoxy composites. COMPOSITES SCIENCE

AND TECHNOLOGY, 55, No. 4, 1995, pp. 405-412.