Laser structural training, artificial intelligence-based acoustic emission localization and structural/noise signal distinguishment in a thick FCEV fuel tank

https://doi.org/10.1016/j.ijhydene.2021.10.262Get rights and content

Highlights

  • Acoustic emission structural health monitoring for FCEV fuel tank.

  • Acoustic emission localization using laser training and artificial intelligence.

  • Collecting four types of unwanted noise from the FCEV fuel tank.

  • Classification of acoustic emission and noise using convolutional neural network.

  • Feature visualization of acoustic emission and noise using artificial intelligence.

Abstract

Acoustic emission (AE) is present when transient elastic waves from structures are generated by various causes, such as structural cracks, fiber breakage, debonding of fibers and matrix, temperature changes, and fatigue. In AE-based structural health monitoring, the simple event counting method is unable to determine where AE occurs, so it is possible to discard a structure even if it is safe or not discard it when it is not safe. Much research on AE localization has been conducted to solve these problems. However, most of the methods have limitations with respect to isotropic material or near field conditions and cannot be applied when there is a change in the boundary conditions of the structure or obstacles. Thus, to solve these problems, a Q-switched laser capable of generating elastic waves has been used to scan and train the structures. Although this method worked effectively on thin specimens, a more advanced method is required for thick and complex structures, such as a fuel tank of a fuel cell electric vehicle (FCEV). Therefore, we propose a novel method based on artificial intelligence (AI) that can be applied to a real FCEV fuel tank fabricated with a filament winding composite. More specifically, this technique modulates the difference in characteristics between AE and laser-induced elastic waves in the frequency domain with AI. AE is simulated by a pencil lead break of the Hsu-Nielsen source. Then, AE localization is performed through cross-correlation in the time⋅frequency domains between a generated AE signal and modulated laser-induced signals obtained from AI. In addition, an experiment conducted to localize the AE that occurs at arbitrary points in real time confirms that AE localization can be performed within 2 s. Finally, an AI algorithm is proposed to distinguish between structural AE and unwanted noise to consider real-world applications and visualize the features of these two types of signals.

Introduction

Acoustic emission (AE) is present when transient elastic waves from structures are generated by various causes, such as structural cracks, fiber breakage, debonding of fibers and matrix, temperature changes, and fatigue. It is important to detect the location of AE occurrence because AE proves that there has been damage to the structure. More specifically, AE propagates from cracks and is collected by a sensor in the form of a surface wave. The method of simply counting the number of AE occurrences throughout a structure and discarding the structure if the number of AE occurrences exceeds the specific criterion, which has been used previously, is not efficient in terms of cost and safety. For example, as shown in Fig. 1, the structure discarding criterion is 10 AE occurrences, and the situations where the structure is discarded because of AE occurrences exceeding the criterion are considered positive situations. In the case of false positives, distributed AEs across the structure are not dangerous, but if the number of AE occurrences exceeds the structure discarding criterion, such as 11 AE occurrences, the structure must be discarded, resulting in wasted cost. Conversely, for false negatives, concentrated AEs are dangerous, but if the number of AE occurrences does not exceed the structure discarding criterion, such as 9 AE occurrences, the structure is not discarded, resulting in safety issues. To avoid the above situations, it is important to determine how many AEs have occurred for each location of the structure or whether there are points where AE accumulates. Therefore, many previous types of research have attempted to find the AE occurrence point, and the AE technology to be introduced in this study is a promising candidate for structural health monitoring (SHM) of fuel cell electric vehicle (FCEV) fuel tanks.

Briefly, introducing the preceding study, the triangulation technique finds the AE occurrence location using the difference in AE signal arrival time of the sensors [1]. The other methods, based on beamforming, select a reference point with 4–8 sensors and find the location of AE based on the delay-and-sum algorithm [2,3]. Also, based on strain rosette technique [[4], [5], [6]], some packets configured with a micro-fiber composite (MFC) sensors in a rosette array are used to obtain vectors in the principal strain direction to find the AE occurrence location. These methods are effective for isotropic materials and near field conditions. However, AE localization results are not accurate in anisotropic materials, such as carbon fiber reinforced plastic (CFRP), because the speeds of the AE signal progressing along each axis of a structure are different. Therefore, Kundu et al. suggested a localization technique in anisotropic plates [7]. Set three or more sensor arrays into one packet, then use the difference of AE signals' arrival time for each sensor to obtain direction vectors indicating the AE occurrence point and then predict the location. Nakatani et al. [8] proposed an improved beamforming method applicable to anisotropic structures. These methods are able to work in quasi-isotropic and anisotropic materials. However, if AE spreads in complex structures with many obstacles, such as holes and boundary condition changes, AE localization becomes difficult. Also, there are limitations in that it was verified only on a thin plate. Additionally, the improved beamforming method has a disadvantage in that it must have signal propagation speed data for each structure direction. To solve these problems, various methods have been suggested. Firstly, Jang et al. [9] proposed the improved triangular method using with magnitude and arrival time of the signals. However, the experiment was conducted only in a thin structure, and there is a disadvantage that a grid must be drawn on the structure to proceed with the experiment. Lastly, the other method based on similarity was largely used. More specifically, the method of AE localization through a similarity check using cross-correlation between the prepared reference data made from laser scanning for elastic wave generation was proposed. The cross-correlation equation is shown in Equation (1). Park et al. [10] obtain reference data in advance using scanning laser Doppler vibrometer (SLDV) and lead zirconate titanate (PZT), and then localize the AE as the point where the highest similarity result appeared. However, this method not only requires many sensors in a small area but also the surface treatment of the structure must be performed to increase the signal-to-noise ratio (SNR), and it has been verified only in thin structures. Also, previous research conducted by Jung et al. [11] used similarity-based method. However, the AE localization technique of previous research was applied only on thin structures less than 3 mm thick, such as aluminum plates and the composite wing skin of an aircraft. Furthermore, the AE localization technique suggested by the previous method did not produce good results in composite structures thicker than 20 mm, such as FCEV fuel tanks, because the mean error of the AE localization exceeded the 10 mm tolerance. Since the previous method created the reference data by selecting one gain from a limited set of gains, it tended to be hard to characterize the generating AE signal at each point on the structure. This tendency is fatal to SHM, which must be done at high speeds. To perform fast SHM, the reference data of the structural points are checked at specific intervals rather than checking all reference data, and then inspecting the area around the most similar structural point once again. Therefore, SHM accuracy is inevitably reduced if there are several reference data areas similar to the generated AE signal or if reference data generation (modulation) is not performed properly. Additionally, Al-Jumaili et al. [12] proposed an AE localization technique using automatic delta-T-mapping (DTM) in a thick structure. However, AE localization takes more than 60 s, which makes it difficult to apply to immediate SHM, and many AE sensors are required in small areas. There is also a limitation in that it has been verified in a small simple steel structure.(fg)(τ)=f(t)g(t+τ)dt,(τ:timedelay)

Since the emergence of compressed natural gas vehicles, dozens of explosion accidents have occurred at home and abroad [13]. Additionally, hydrogen explosions are very dangerous. Therefore, much research has been conducted to prevent explosions in places that use hydrogen, such as hydrogen fueling stations [14]. Additionally, various studies on safe hydrogen fuel storage containers are being conducted. More specifically, research on container manufacturing methods or materials is being conducted to prevent accidents in hydrogen fuel storage containers [15,16]. According to the study by Tanç et al. [17], infrastructure related to battery electric vehicle (BEV) and FCEV will increase by 2050. Also, as technology advances, it is predicted that FCEV will be commercialized and its demand will increase. In this environment, explosion accidents of FCEV fuel tanks containing pressurized hydrogen are sensitive matters because they directly relate to human safety. Additionally, there is no guarantee that FCEV accidents will not happen. Therefore, detecting AE occurrence locations for the safety of sensitive structures is essential for SHM and can help prevent accidents.

Therefore, this study is mainly focused on an FCEV fuel tank with a thickness thicker than of 20 mm to consider safety issues and overcome the limitations of the AE localization technique on the thick structure shown in previous research [11,12] by utilizing artificial intelligence (AI) algorithms. Also, the similarity-based AE localization method, which can be universally used in anisotropic structures, is applied to complex structures (composite stiffened curved panel) and complex thick structures (FCEV fuel tank), respectively. Additionally, as part of the verification of the AE localization results improvement, the advanced AE localization method using AI is applied to a composite stiffened curved panel as well as the FCEV fuel tank and compared with the results of previous research methods [11]. Finally, in the actual situation where the FCEV fuel tank is installed, unwanted noise other than the actual AE can be collected by AE sensors. Noise can be generated from sensors, external noise, and FCEV tank joint parts, and thus selective collection of AE and noise signals is needed. This paper also develops an AI algorithm for AE and noise signal distinguishment and visualizes the features of these two types of signals.

Section Fuel cell electric vehicle fuel tank describes the specifications of the FCEV fuel tank used in the experiment. Section Data acquisition explains laser-induced elastic wave and AE signal acquisition methods obtained by a Q-switched laser-based system and pencil lead break (PLB), respectively, on an FCEV fuel tank. Section Signal modulation and acoustic emission localization introduces and mentions the differences between the laser-induced elastic wave modulation algorithm used in previous research [11] and the AI-based modulation algorithm developed in this research. Section Noise distinguishment algorithm shows AE and different types of noise that are able to occur in the FCEV fuel tank and briefly explains these two types of signal distinguishing algorithms. Section Acoustic emission localization in the composite stiffened curved panel compares the results of AE localization using two methods described in Section Signal modulation and acoustic emission localization for the composite stiffened curved panel used in the previous study [11]. These results prove that the method proposed in this study is applicable to complex structures and produce better results compared to previous method. Then, Section Acoustic emission localization in the FCEV fuel tank compares the results of AE localization using the two methods for the FCEV fuel tank and analyzes the results with cross-correlation maps. These results show that AI-based methods are applicable in complex thick structures unlike previous method [11]. Section Acoustic emission localization in real-time in the FCEV fuel tank describes the real-time AE localization experiment that must be performed quickly. Using the method proposed in this section, AE localization was possible within 2 s for an area of 1,200 × 200 mm2, and the limitations of previous studies [12] have been overcome. Section Comparison of acoustic emission and unwanted noise distinguishment results by AI algorithms explains three different conditions of AI algorithms to compare the results of the AE and noise signal distinguishment and visualizes these two types of signal feature results. These results increase the applicability of the technique by filtering out the noises that may occur in real situations where the structure is driven. Finally, Section Conclusion concludes this paper.

Section snippets

Fuel cell electric vehicle fuel tank

This article mainly focuses on developing an AE technique-based SHM system for FCEV fuel tanks, as shown in Fig. 2. The FCEV fuel tank is a type IV that is fabricated by the filament winding process. In this study, only part of the structure was wound with GFRP to create a complex situation that may occur in the FCEV fuel tank. Details for dimensions are shown in Fig. 2.

Data acquisition

For the reasons mentioned in Section Introduction, reference data must be collected in advance, and when an AE occurs, the

Acoustic emission localization in the composite stiffened curved panel

Before showing the localization results of the FCEV fuel tank, the FESM and the GAN-based method explained in Section Signal modulation and acoustic emission localization were applied to a stiffened panel of 3 mm thickness to compare results. The slightly curved panel made of CFRP has substructures such as stringers and spars, as shown in Fig. 15. Fig. 15(a) shows the backside of the composite stiffened curved panel and attached substructures, and Fig. 15(b) shows the location of the

Conclusion

In this paper, the technique of AE localization based on a Q-switched laser with a single channeled three PZT AE sensor and AI has been proposed for the FCEV fuel tank. In addition, an algorithm to distinguish between structural AE signals and unwanted noise signals using AI has been proposed, taking into consideration the possibility of using the structure in the real world. Regarding the signal acquisition, reference data for AE localization were collected through laser scanning of the FCEV

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work was supported by the National Research Foundation of Korea(NRF) Grant funded by the Ministry of Science and ICT (NRF-2017R1A5A1015311).

References (23)

  • D.C. Betz et al.

    Structural damage location with fiber Bragg grating rosettes and lamb waves

    Int J Struct Health Monit

    (2007)
  • Cited by (0)

    View full text