a r t i c l e i n f o a b s t r a c t
Article history: A time–frequency technique for locating leaks in buried gas distribution pipes involves the use of the
Received 10 August 2007 cross-correlation on two measured acoustic signals on either side of a leak. This technique can be
Received in revised form problematic for locating leaks in steel pipes, as the acoustic signals in these pipes are generally
7 August 2008
narrow-band and low frequency. The effectiveness of the time–frequency technique for detecting leaks
Accepted 7 August 2008
in steel pipes was investigated experimentally in an earlier study. The object of this paper is to identify
the characteristics of this dispersive acoustic wave through analysis of the cut-off frequency by using
Keywords:
the time–frequency method experimentally and BEM (boundary element method) theoretically for the
Leakage
Gas duct development of an experimental tool to analyze the leak signals in steel pipe. The tool is based on
Detection experimental work and theoretical formulation of wave propagation in a fluid-filled pipe. This tool uses
Time–frequency method the time–frequency method to explain some of the features of wave propagation measurements made
Acoustic wave in gas pipes. Leak noise signals are generally passed through a time–frequency filter for detection of
impulse signal related leakage.
Ó 2008 Elsevier Ltd. All rights reserved.
0950-4230/$ – see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jlp.2008.08.009
M.-S. Kim, S.-K. Lee / Journal of Loss Prevention in the Process Industries 22 (2009) 990–994 991
Table 1
kr k Theoretical cut-off frequencies of a steel pipe
a r
n Cut-off frequencies: fm,n (Hz)
kz z
m
0 1 2 3 4
0 0 1747.6 2896.9 3989.2 5053.0
Fig. 1. Boundary condition of a cylindrical duct.
1 3637.8 5062.5 6373.3 7617.6 8814.3
2 667.7 8111.5 9469.7 10,780.5 12,043.8
3 9659.7 11,122.4 12,509.2 13,857.9 15,159.2
4 12,651.6 14,114.4 15,529.6 16,897.4 18,236.6
For cylindrical coordinates with boundary condition as shown in
Fig. 1, the solution Eq. (2) is rewritten by,
4m;n ðt; rÞ ¼ 4f ;m;n ðrÞexp þ j2pfm;n t ; m; n ¼ 0; 1; 2; . (4)
4f ;m;n ðr; j; zÞ ¼ Jm kr;m;n r Am;n cosðmjÞ þ Bm;n sinðmjÞ
eðjkz zÞ ; m; n ¼ 0; 1; 2. ð5Þ
where 4f(r) is the only space vector. The 4f(r) is the acoustic velocity
potential and depends on only the space vector. The m and n are
integers. In Eq. (5), from the boundary condition in the radial
direction, the particle velocity becomes as follows:
Fig. 3. Drawing of the test duct.
uf ða; j; zÞ ¼ V4f ða; j; zÞ ¼ 0 (6)
In case of the cut-off frequency f ¼ fm,n and kz,m,n ¼ 0, the wave is
Using this boundary condition Eq. (6), the Bessel function in Eq. not propagated to the z-direction. Therefore, the wave number in
(5) is rewritten by, the z-direction kz,m,n depends on the acoustic wave frequency and
0
there are two considered cases.
Jm kr;m;n a ¼ 0; m; n ¼ 0; 1; 2; . (7)
Fig. 2. Acoustic cavity mode of a duct. Fig. 4. Experimental set-up of the test duct.
992 M.-S. Kim, S.-K. Lee / Journal of Loss Prevention in the Process Industries 22 (2009) 990–994
Table 2
Sensor details used for a test
Ch. 1 Ch. 2
Type Pressure Pressure
Model B&K 4935 B&K 4935
Label Array Mic.4935 Array Mic.4935
Sensitivity 5.281e-03 V/Pa 6.756e-03 V/Pa
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
8
< k 1 fm;n =f ; f fm;n ; Propagation mode
kz;m;n ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 (10)
:
jk 1 fm;n =f ; f fm;n ; Evancescent mode
Fig. 5. Time vs frequency plot by STFT (short time Fourier transform) method (a)
sensor 1, (b) sensor 2. Fig. 7. Model of BEM analysis (a) mesh model (b) acoustic field point.
M.-S. Kim, S.-K. Lee / Journal of Loss Prevention in the Process Industries 22 (2009) 990–994 993
are inserted into the duct to detect the acoustic wave. The detailed
experimental set-up for the duct is shown in Fig. 4 and the sensors
used for the test are listed in Table 2. Fig. 5 shows the image map for
the STFT (short time Fourier transform) of two signals measured at
sensor 1 and sensor 2. According to these results, we can see the
cut-off frequencies around 1938 Hz, 3162 Hz and 3988 Hz. These
frequencies correspond to the cut-off frequencies for acoustic
modes (1,0), (2,0) and (0,1), respectively, listed in Table 1. There is
a little difference between the experimental results and theoretical
results since the physical parameters are not exactly the same. The
spectrum component at low frequencies is due to the reflection of
the plane wave inside a duct. Fig. 6 shows the power spectral
density (PSD) for both signals measured at sensor 1 and sensor 2.
From these results, it is difficult to see the dispersive phenomenon
and to find cut-off frequencies. Therefore, the time–frequency
4. Theoretical simulation
Fig. 8. Three low acoustic modes of duct (a) (1,0) mode (b) (2,0) mode (c) (0,1) mode. Fig. 10. Group and phase speed difference between three methods.
994 M.-S. Kim, S.-K. Lee / Journal of Loss Prevention in the Process Industries 22 (2009) 990–994