94
B. P. Duong et al.
of the pipelines and on-time maintenance have inspired the reconsideration over
the recent years. Numerous diagnosis programs can early recognize some abnormal
symptoms in the pipeline system, including hardware-based methods, fiber optic,
visual inspection, cable sensor, and acoustic emission (AE) [2–4]. Apparently, the
AE signal techniques have been auspicious since it exhibited effectively for offering
high sensitivity concerning fault detection, quickly recognize small leaks, and supply
the advances in data collection and analyzing speeds with the huge amount of gathered signals [5, 6]. Pipeline leakage mechanism creates the high-frequency signal
in the elastic wave types which transmitted along the tube wall, reflected between
the two surfaces of the pipeline [7]. When a solid wall of the pipe is excited by an
abrupt displacement or an impact, it creates a dynamic vibration problem (generate
dynamic stress). Escaping liquid in high-pressure pipes passes through a perforation,
releases elastic energy in the form of stress waves caused by localized loss of energy,
and these generated signals usually represent the abnormal AE events. Stress waves
are transmitted through the pipe wall which can be collected by using an acoustic
emission sensor installed on the pipe body [8, 9]. The analysis of the AE event from
leakage obtained via the pipe body stimulation is a significant key for leakage fault
diagnosis. However, the signal characteristics that are non-stationary could be treated
as a broadband random noise modulated by a slow-changing random process and
are ineliminable contamination with non-leak signals that make it difficult to distinguish AE signals. The AE event that makes an appearance from an abrupt change
in flow generates transient bursts. By detecting a burst event, it is also essential to
compute the time-differences-of-arrival (TDOA) between AE signals recorded by the
sensors, which aids for AE source localization. Therefore, burst segmentation and
exact analysis of the characteristics of this AE wave have acquired useful knowledge. In previous research, some researchers have been introduced to enhance leak
detection in water-filled pipelines using the wavelet decomposition and denoising
has to recognize the transient bursts [10]. However, the presented methods have their
own limitation because most of the components decomposed wavelet transform are
mostly related to the remains of attenuated leak signals. On the other hand, an important issue existing in the fault diagnosis algorithm is to distinguish the leakage signal
without giving false alarms. Previously, object detection with a constant false alarm
rate (CFAR) was employed in the radar system with the doppler effect [11]. However,
the considering of burst inside the AE leak signal as objects, which are necessary to
be identified and employ the detector with CFAR, has not been researched yet.
The main objective of this research is detection the burst events associated with
leakage and isolation of the burst waveform by the Enhance Constant Faults Alarm
Rate (ECFAR) methodology to identify the characteristics of this dispersive AE
signal in leakage phenomenon. Then, the investigating characteristics of bursts are
utilized to distinguish the different levels of leak signal using the multiclass support
vector machines in the strategy of one-against-all (OAA-MCSVM).
The organization of this paper is illustrated as follows. The data acquisition system
to record the leakage dataset in the type of acoustic emission, which is applied with
the proposed algorithm for evaluation purposes and the testbed are introduced in
Sect. 9.2. After that, Sect. 9.3 describes in detail burst event detection and isolation
B. P. Duong et al.
of the pipelines and on-time maintenance have inspired the reconsideration over
the recent years. Numerous diagnosis programs can early recognize some abnormal
symptoms in the pipeline system, including hardware-based methods, fiber optic,
visual inspection, cable sensor, and acoustic emission (AE) [2–4]. Apparently, the
AE signal techniques have been auspicious since it exhibited effectively for offering
high sensitivity concerning fault detection, quickly recognize small leaks, and supply
the advances in data collection and analyzing speeds with the huge amount of gathered signals [5, 6]. Pipeline leakage mechanism creates the high-frequency signal
in the elastic wave types which transmitted along the tube wall, reflected between
the two surfaces of the pipeline [7]. When a solid wall of the pipe is excited by an
abrupt displacement or an impact, it creates a dynamic vibration problem (generate
dynamic stress). Escaping liquid in high-pressure pipes passes through a perforation,
releases elastic energy in the form of stress waves caused by localized loss of energy,
and these generated signals usually represent the abnormal AE events. Stress waves
are transmitted through the pipe wall which can be collected by using an acoustic
emission sensor installed on the pipe body [8, 9]. The analysis of the AE event from
leakage obtained via the pipe body stimulation is a significant key for leakage fault
diagnosis. However, the signal characteristics that are non-stationary could be treated
as a broadband random noise modulated by a slow-changing random process and
are ineliminable contamination with non-leak signals that make it difficult to distinguish AE signals. The AE event that makes an appearance from an abrupt change
in flow generates transient bursts. By detecting a burst event, it is also essential to
compute the time-differences-of-arrival (TDOA) between AE signals recorded by the
sensors, which aids for AE source localization. Therefore, burst segmentation and
exact analysis of the characteristics of this AE wave have acquired useful knowledge. In previous research, some researchers have been introduced to enhance leak
detection in water-filled pipelines using the wavelet decomposition and denoising
has to recognize the transient bursts [10]. However, the presented methods have their
own limitation because most of the components decomposed wavelet transform are
mostly related to the remains of attenuated leak signals. On the other hand, an important issue existing in the fault diagnosis algorithm is to distinguish the leakage signal
without giving false alarms. Previously, object detection with a constant false alarm
rate (CFAR) was employed in the radar system with the doppler effect [11]. However,
the considering of burst inside the AE leak signal as objects, which are necessary to
be identified and employ the detector with CFAR, has not been researched yet.
The main objective of this research is detection the burst events associated with
leakage and isolation of the burst waveform by the Enhance Constant Faults Alarm
Rate (ECFAR) methodology to identify the characteristics of this dispersive AE
signal in leakage phenomenon. Then, the investigating characteristics of bursts are
utilized to distinguish the different levels of leak signal using the multiclass support
vector machines in the strategy of one-against-all (OAA-MCSVM).
The organization of this paper is illustrated as follows. The data acquisition system
to record the leakage dataset in the type of acoustic emission, which is applied with
the proposed algorithm for evaluation purposes and the testbed are introduced in
Sect. 9.2. After that, Sect. 9.3 describes in detail burst event detection and isolation