80
L. L. Ting et al.
of NRW are due to leakages, pipe bursts, public usage, metering errors and theft, but
it has been identified as mainly due to to leakages [2].
Leaks result in serious consequences such as loss of money and natural resources.
They also create public health risk because contaminants may pass into the pipe when
there is a pressure drop in the pipe [2]. The critical aftermath caused by leakages
is the unnecessary water wastage. This may put up to water demand along with
population growth and urbanization [3]. For this reason, water leakage control has
been a growing concern in many countries with an aim to promote water conservation.
Detection of water leaks using acoustic method is a popular way because it can
provide a faster and real-time response [4] in spite of the occurrence of false alarm
or missed detection is high [5]. According to the ASTM E1316 Standard, Acoustic
Emission (AE) is a phenomenon by which emission of transient waves due to the
rapid release of energy in a material. When there are discontinuities in the material,
it will create sound and the sound can be captured by the sensors. Thereafter, the
leak source can be located by means of determining the time difference between the
sound waves reach sensors.
Essentially, the signals must be pre-processed first in order to filter noise out.
Noise is unwanted signals that inevitably exist in the signals. In the work conducted
by Liu [4], the leakage signal is prominent after filtering. Therefore, noise reduction
is a crucial step in acoustic leak detection method. Digital filtering is generally used to
eliminate noise from an interested signal if the noise bandwidth is known. Commonly
defined filter types are low pass, high pass, band pass and band stop. Hunaidi [2]
and Srirangarajan et al. [6] applied low pass and band pass filtering in their leak
detection practices to band-limit the signals. However, design of filter requires prior
information of both signal and noise so it is only applicable when desired signal has
different frequency with the noise [4].
On the other hand, wavelet de-noising is a noise reduction method through wavelet
analysis. It decomposes a signal into wavelet coefficients and applies a threshold on
the decomposed coefficients in order to eliminate noisy parts. Meng et al. [7] and
Wan et al. [8] employed wavelet de-noising to de-noise their collected signal. Their
results depict that wavelet de-noised signals are effective to increase leak positioning
accuracy by detecting a leak with error of 5%. Nevertheless, there are few indispensable challenges associated with wavelet de-noising which are the decomposition
level [9], threshold method [8] and the threshold value [10].
Several methods had been introduced to extract the feature of non-stationary
leakage signals such as Discrete Wavelet Transform (DWT) [9], Wavelet Packet
Decomposition (WPD) [11, 12], Empirical Mode Decomposition (EMD) [13, 14] and
Dual-Tree Complex Wavelet Transform. DTCWT is introduced in this work because
it is a propitious method to restrain acoustic noise without eliminating useful signal
characteristics [15]. The demerits of DWT and WPD are lack of shift-invariance,
poor directional selectivity and frequency aliasing. Diversely, EMD suffers from
major difficulty such as mode mixing because it is difficult to decompose a signal
that has multiple overlapping modes [16].
The concept of leak localization is illustrated in Fig. 8.1. The waves due to leak
will travel in both upstream and downstream of the pipe. Consequently, the waves
L. L. Ting et al.
of NRW are due to leakages, pipe bursts, public usage, metering errors and theft, but
it has been identified as mainly due to to leakages [2].
Leaks result in serious consequences such as loss of money and natural resources.
They also create public health risk because contaminants may pass into the pipe when
there is a pressure drop in the pipe [2]. The critical aftermath caused by leakages
is the unnecessary water wastage. This may put up to water demand along with
population growth and urbanization [3]. For this reason, water leakage control has
been a growing concern in many countries with an aim to promote water conservation.
Detection of water leaks using acoustic method is a popular way because it can
provide a faster and real-time response [4] in spite of the occurrence of false alarm
or missed detection is high [5]. According to the ASTM E1316 Standard, Acoustic
Emission (AE) is a phenomenon by which emission of transient waves due to the
rapid release of energy in a material. When there are discontinuities in the material,
it will create sound and the sound can be captured by the sensors. Thereafter, the
leak source can be located by means of determining the time difference between the
sound waves reach sensors.
Essentially, the signals must be pre-processed first in order to filter noise out.
Noise is unwanted signals that inevitably exist in the signals. In the work conducted
by Liu [4], the leakage signal is prominent after filtering. Therefore, noise reduction
is a crucial step in acoustic leak detection method. Digital filtering is generally used to
eliminate noise from an interested signal if the noise bandwidth is known. Commonly
defined filter types are low pass, high pass, band pass and band stop. Hunaidi [2]
and Srirangarajan et al. [6] applied low pass and band pass filtering in their leak
detection practices to band-limit the signals. However, design of filter requires prior
information of both signal and noise so it is only applicable when desired signal has
different frequency with the noise [4].
On the other hand, wavelet de-noising is a noise reduction method through wavelet
analysis. It decomposes a signal into wavelet coefficients and applies a threshold on
the decomposed coefficients in order to eliminate noisy parts. Meng et al. [7] and
Wan et al. [8] employed wavelet de-noising to de-noise their collected signal. Their
results depict that wavelet de-noised signals are effective to increase leak positioning
accuracy by detecting a leak with error of 5%. Nevertheless, there are few indispensable challenges associated with wavelet de-noising which are the decomposition
level [9], threshold method [8] and the threshold value [10].
Several methods had been introduced to extract the feature of non-stationary
leakage signals such as Discrete Wavelet Transform (DWT) [9], Wavelet Packet
Decomposition (WPD) [11, 12], Empirical Mode Decomposition (EMD) [13, 14] and
Dual-Tree Complex Wavelet Transform. DTCWT is introduced in this work because
it is a propitious method to restrain acoustic noise without eliminating useful signal
characteristics [15]. The demerits of DWT and WPD are lack of shift-invariance,
poor directional selectivity and frequency aliasing. Diversely, EMD suffers from
major difficulty such as mode mixing because it is difficult to decompose a signal
that has multiple overlapping modes [16].
The concept of leak localization is illustrated in Fig. 8.1. The waves due to leak
will travel in both upstream and downstream of the pipe. Consequently, the waves