Professional Documents
Culture Documents
Sharp
Abstract
This paper discusses the measurement of the acoustical properties of tubular systems
using pulse reflectometry. The technique involves the injection of a sound pulse into
the object under investigation and the recording of the resultant reflections. Analysis
of the reflections gives information about the bore profile and input impedance of the
object. The effect of small leaks on both of these properties is investigated, leading
to a method for detecting the presence of a leak in a general tubular system. For the
special case of a cylindrical pipe, the position and size of the leak are also determined.
1 Introduction
The detection of leaks is of crucial importance in many industrial situations. One of
the standard methods of testing that a pipe is airtight involves sealing the end of the
pipe, increasing the air pressure within to a certain level and confirming that this level
remains constant. However, in industry, many systems are inaccessible and impossible
to seal, so in these cases a non-invasive technique is necessary. Here, an alternative
method of leak detection, based on the technique of acoustic pulse reflectometry, is
presented.
Originally developed for use in seismic studies [Ware & Aki, 1969], pulse reflec-
tometry has more recently been applied to the investigation of ducts of varying cross-
section. In the medical field, extensive research has been carried out into the use of the
technique for the measurement of airway dimensions [Jackson et al, 1977; Fredberg et
al, 1980; Brooks et al, 1984; Marshall, 1992]. At present, the main focus of the work
being carried out on ducts is in the field of musical instrument research. Pulse reflec-
tometry enables the measurement of the input impulse response of a wind instrument
[Smith, 1988; Watson & Bowsher, 1988]. Using a suitable reconstruction algorithm, a
reconstruction of the internal bore profile of the instrument can be evaluated from the
input impulse response. The input impedance of the instrument can also be calculated
[Watson, 1989].
The present paper is concerned with the detection of leaks in tubular systems. The
technique of pulse reflectometry is described, discussing in detail the measurement of
2 D.B.Sharp
the input impulse response of a general tubular object with varying cross-section. From
the input impulse response, an internal bore reconstruction and an input impedance
curve of the object can be evaluated. The increased accuracy of bore measurement
obtained by using the loss-compensating reconstruction algorithm recently proposed
by Amir, Rosenhouse and Shimony [1995] has made it possible to develop a technique
for the detection of small leaks in tubular systems. For the special case of a cylindrical
pipe, it is demonstrated that the axial position of a side hole can be identified and its
size evaluated.
2 Basic Technique
2.1 Determination of the input impulse response
The technique of pulse reflectometry involves injecting a pulse of sound into a tubular
object and digitally recording the resultant reflections. Analysis of these reflections
provides information about the internal bore of the object.
COMPUTER
D/A A/D
AMPLIFIER
AMPLIFIER
& FILTER
MICROPHONE
l1 l2
Figure 1 shows a schematic diagram of the pulse reflectometer used in the present
study. The reflectometer and test object are mounted in an anechoic chamber, with the
electronics in an adjoining control room. An electrical pulse (of width 80µs and voltage
5V) is produced by a 12 bit D/A converter (located on an Iotech DaqBoard 100A
data acquisition board inside a Viglen 486DX 66MHz PC). The pulse is amplified
3 D.B.Sharp
by a Pioneer A-119 stereo amplifier, and used to drive a Fane Professional MD2050
compression driver loudspeaker. The resultant sound pressure pulse travels along a 6m
long copper source tube (of internal radius 4.8mm and wall thickness 1.2mm) which
is clamped to a wooden board in a spiral of approximately 200mm radius. A Knowles
microphone embedded halfway along the tube records the reflections returning from
the tubular object under test, which is coupled to the far end of the source tube. The
microphone output is amplified by a second amplifier and low-pass filtered (using a
Barr and Stroud EF4-03 filter set to 20kHz) to prevent aliasing. The resultant signal is
then sampled by a 12 bit A/D converter (using a sampling frequency of 50kHz and a
sample length of 1024 points, giving a sample time of approximately 20ms) and stored
on the PC. The A/D converter is located on the DaqBoard 100A data acquisition board.
This procedure is repeated 1000 times and the samples are averaged to improve
the signal-to-noise ratio. Precise time alignment of successive samples is achieved by
triggering the sampler from the electrical pulse to the loudspeaker. A delay of 180ms
is included before each repetition to ensure all the signal from the previous step has
died away.
The source tube half-length l2=3m is necessary to ensure that the input pulse has
fully passed the microphone before the first of the returning object reflections reaches
it. The minimum duration of the input pulse is in practice limited by the requirement
that the pulse carries sufficient energy to ensure a good signal-to-noise ratio in the
measured reflections.
After the object reflections pass the microphone they are further reflected by the
loudspeaker. The source tube half-length l1 =3m is necessary to separate the object
reflections from these source reflections. The object reflections must be sampled over
a time period no longer than the time taken to travel the distance 2l1, to ensure that no
source reflections are recorded. This restricts the maximum length of objects that can
be measured to 3m in the present case.
For an ideal delta function sound pressure pulse, the reflections obtained from the
tubular object under test would be its input impulse response. However, the sound
pressure pulse is not ideal; to obtain the input impulse response, the reflections are
deconvolved with the input pulse shape. The input pulse shape is measured by termi-
nating the source tube with a flat perspex plate of thickness 5mm and recording the
reflected pulse (so taking losses along the source tube into consideration) [Sondhi &
Resnick, 1983].
The deconvolution is carried out by Fourier transforming both the sample contain-
ing the object reflections and the sample containing the input pulse (each of length
1024 points). A complex division of the object reflections by the input pulse is then
carried out in the frequency domain. A constraining factor q is added to the denomi-
nator to prevent division by zero [Marshall, 1990]:
R ω I ω
IIR ω (1)
I ω I ω q
where R ω is the transformed object reflections, I ω is the transformed input pulse,
4 D.B.Sharp
2.2.1 Results
Figure 2 shows the internal profile of a 356mm long stepped tube (whose radius ex-
pands from 4.7mm to 6.2mm to 9.25mm), reconstructed using the input calibration
method of DC offset evaluation. Note the presence of the 403mm long connector at
the start of the reconstruction.
The overshoots at each expansion in the reconstruction are due to the presence
of ripple in the input impulse response. This ripple, an example of the Gibbs phe-
nomenon, is introduced when the input pulse and the reflections are Fourier trans-
formed, because of the finite bandwidth of the transformed spectra.
5 D.B.Sharp
10.0
PULSE REFLECTOMETER
DIRECT MEASUREMENT
9.0
8.0
radius/mm
7.0
6.0
5.0
4.0
0.0 0.2 0.4 0.6 0.8
distance/m
Figure 2: Bore reconstruction of stepped tube
radius/mm
-5
5
0
-5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
distance/m
Figure 3 shows a 3D display of the same stepped tube. This time, the connector is
not displayed.
p t
δ t
iir t
(2)
zst u t
δ t
iir t
(3)
where p t
is the pressure recorded by the microphone at time t, deconvolved with the
input pulse shape, u t
is the volume velocity at the microphone at time t, zst ρc S is
the acoustic impedance of the source tube (ρ is the air density, c is the speed of sound
in air, S is the cross-sectional area of the source tube) and iir t
is the input impulse
response of the object.
In the frequency domain this gives
P ω
1 IIR ω
(4)
zst U ω
1 IIR ω
(5)
where P ω
is the Fourier transform of p t
, U ω
is the Fourier transform of u t
and
IIR ω
is the Fourier transform of iir t
.
Hence, the input impedance is given by:
P ω
1 IIR ω
z ω
zst (6)
U ω
1 IIR ω
in
In the evaluation of the input impedance, it is not necessary to calculate the DC offset
in the input impulse response.
2.3.1 Results
Figure 4 compares impedance measurements using pulse reflectometry (diamonds)
with an impedance curve measured using a conventional frequency domain technique
(both curves are for the stepped tube displayed in figures 2 and 3). In the conventional
technique, a sine wave is passed down a high impedance capillary into the object being
measured. The high capillary impedance provides an approximately constant volume
7 D.B.Sharp
8
10
impedance/ohms
7
10
6
10
PULSE REFLECTOMETER
SWEPT SINE WAVE
frequency/kHz
Figure 4: Impedance curve for stepped tube
velocity which ensures that, to the same approximation, the pressure amplitude is pro-
portional to the input impedance. The pressure amplitude is measured by a microphone
and converted to the input impedance by dividing by the volume velocity. This is re-
peated for a whole range of sine wave frequencies [Backus, 1974; Campbell, 1994].
The frequency resolution of the impedance measurement using pulse reflectom-
etry is constrained by the sample length of the object reflections, which is in turn
constrained by the source tube length l1 . For l1 =3m, the maximum sample length is
approximately 20ms, leading to a frequency resolution of approximately 50Hz. To
improve the resolution by sampling over a longer period of time either the source tube
length l1 must be increased or the source reflections removed. Although increasing
the source tube length improves the resolution of the impedance measurement, the
signal experiences extra losses. This attenuation of the higher frequencies means that
the impedance can only be accurately calculated at low frequencies. At present, work
is being carried out on removing the source reflections (by driving the loudspeaker
in such a way as to cancel out the incoming object reflections). This should result
in improved resolution without reducing the range over which the impedance can be
measured.
8 D.B.Sharp
3.1 Results
Figure 5 shows bore reconstructions of a 501mm long cylindrical pipe (of internal
radius 5mm and wall thickness 1mm) both with and without a 0.5mm radius sidehole.
Again, note the presence of the 403mm long connector at the start of the reconstruction.
The expansion in the reconstruction of the cylindrical pipe with the hole is clear.
For the case of a cylinder, it is quite obvious that this expansion is caused by the leak
because the radius at the end of the pipe reconstruction should be equal to the radius
at the beginning. The position of the leak, 301mm from the output end of the pipe, is
also clear.
For an object whose bore profile is not known in advance, an expansion in the re-
construction due to a leak may be indistinguishable from an actual widening of the
bore. Hence, the position of a leak may be difficult to determine. However, the pres-
ence of the leak can still be confirmed by observing a discrepancy between the radius
of the reconstructed bore and the directly measured radius at the output end of the tube.
10.0
8.0
radius/mm
6.0
4.0
CYLINDER WITHOUT HOLE
CYLINDER WITH HOLE
2.0
0.0 0.2 0.4 0.6 0.8 1.0
distance/m
Figure 5: Bore reconstruction of cylindrical pipe with and without sidehole
r
z input z load
zload kπr2
j tan kl
ρω ρω
zinput
kπr2 (7)
zload kπr2
1 j tankl
ρω
where zinput is the input impedance, zload is the load impedance (at the end of the cylin-
der), ρ is the air density, ω is the angular frequency, and k is the complex propagation
constant (k k jα).
1 ηω κω
k ω c and α γ 1$
rc ! 2ρ #" ! 2ρc p %
γ is the ratio of the principal specific heats of air
c p is the specific heat of air at constant pressure
η is the coefficient of shear viscosity of air
κ is the thermal conductivity of air
c is the speed of sound in air
zh lh
r
z0 z1 z2 z2rad
l1 rh l2
CYLINDRICAL CYLINDRICAL
SECTION 1 SECTION 2
Thus
z1 z2
zh & (9)
z2 ' z1
4.2.2 Evaluation of z1
Equation 7 gives the input impedance of a cylinder with a complex load impedance
at its end. The first cylindrical section, of length l1, is such a cylinder, with an input
impedance z0 and a complex load impedance z1 . By substituting z0 for zinput , z1 for
zload , l1 for l and rearranging, an expression for the load impedance (i.e. the impedance
at the end of the first cylindrical section, z1 ) in terms of the measured input impedance,
z0 , is obtained:
ρω
z0 ' j tankl1
kπr2
z1 & (10)
z0 kπr2
1' j tan kl1
ρω
4.2.3 Evaluation of z2
Similarly, the impedance of the second cylindrical section can be calculated using
equation 7. The impedance at the input to this cylinder is z2 and, as the cylinder is
open-ended, the load impedance is the radiation impedance, z2rad . Substituting the
relevant values into equation 7 gives:
() ,-
)* z2rad kπr2 -
j tankl2
ρω ρω +
z2 & . (11)
kπr2 z2rad kπr2
1 j tankl2
+ ρω
In this case, the end is unflanged and the radiation impedance is given by [Kinsler
et al, 1982]:
1 ρω 2 2 ρω
z2rad & k r j0 / 6 kr (12)
4 kπr 2 + kπr2
so,
ρω 4 k r + j 1 0 / 6kr + tan kl2 2
1 2 2
z2 &
kπr2 0 1 1 ' 0 / 6kr tankl2 2
(13)
j 14 k2 r2 tankl2 3
+
4.2.4 Evaluation of zh
Substituting equations 10 and 13 into equation 9 gives the complex impedance of the
hole, zh , in terms of the measured impedance at the input to the leaking cylinder, z0 .
12 D.B.Sharp
ρωk ρω ηω κω ρω rh3
zh 4 9 j 6 γ 9 17 8 9 8
r2 A
j l 1 595r 0 58
4πrh c <?= 2ρ 5 = 2ρc p > 5 πrh2 @ 5
h h
4π
(15)
Considering only the imaginary part of the impedance zh :
ρωlh 1 8 595ρω 0 8 58ρωrh ρω ηω κω
zhimag 4 9 9 6 γ 9 17 (16)
πrh 5
2 πrh πr 2 4πrh c <= 2ρ 5 = 2ρc p >
Rearranging gives a cubic equation for rh :
1 8 724πr2 zhimag 2
rh3 r
5CB ρω D h
0 8 431r2 ηω κω
6 γ 9 17 9 2 8 75r2 rh 9FE 1 8 724lhr2 G 4 0 (17)
5 B c < = 2ρ 5 = 2ρc p > D
Substituting the value for zhimag (calculated in section 4.2 from the measured input
impedance) into equation 17 and solving [Press et al, 1988], yields a value for the hole
radius.
The success of equation 17 in predicting the radius of the hole will clearly depend
on a suitable choice of frequency. A small hole has little effect on the impedance of
the air column if it is in the vicinity of a pressure node; thus, at frequencies which
correspond to this condition the prediction of hole size from impedance measurement
can be expected to break down. Likewise, the hole has little effect on the acoustical
properties of the pipe at frequencies much above the cutoff frequency, which is deter-
mined by the radii of the main cylinder and the hole. This sets an upper limit on the
usable frequency range.
13 D.B.Sharp
4.3.1 Results
7.0
REFLECTOMETRY (1.0mm)
REFLECTOMETRY (1.5mm)
6.0 REFLECTOMETRY (2.5mm)
DIRECT MEASURE (1.0mm)
DIRECT MEASURE (1.5mm)
5.0
hole radius/mm
4.0
H
3.0
2.0
1.0
0.0
0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2
frequency/kHz
Figure 8: Comparison of predicted and measured hole radii
Figure 8 illustrates the predictions of equation 17 for three different sizes of hole
in a cylindrical pipe of length 501mm, internal radius 5mm and wall thickness 1mm.
In each case, the hole is 301mm from the output end of the pipe. Graphing the pre-
dicted radius as a function of frequency in this way is useful, since the divergences of
the prediction allow an immediate identification of the frequencies for which pressure
nodes coincide with the hole. In the present case, the divergences occur at integer mul-
tiples of 570Hz, as expected from the known geometry. The predictions also start to
break down, although less dramatically, above approximately 2000Hz, which is in the
vicinity of the cutoff frequency for the largest hole.
As a prescription for selecting valid data for predicting the effective radius of a leak
in an open ended pipe measured in this way, it is suggested that the frequency F of the
first divergence is estimated by inspection of the graph. Values of rh in the frequency
ranges 1 I 25F J 1 I 75F and 2 I 25F J 2 I 75F are then averaged to give a best prediction
of the effective radius of the hole. Predictions drawn in this way from Figure 8 are
compared in Table 1 with direct measurements using calipers. Six predicted values are
averaged in each frequency range; the uncertainties quoted are single standard devia-
tions. For each of the holes, the predicted radius agrees with that measured directly
within experimental error.
14 D.B.Sharp
Acknowledgments
This work has been supported by an EPSRC research studentship. The authors are
grateful to Prof.C.Greated and Dr.R.Parks for help in the early stages of the project;
Dr.N.Amir provided valuable assistance, including the provision of the reconstruction
algorithm prior to publication.
References
AMIR N., ROSENHOUSE G., SHIMONY, U. (1995) “A discrete model for tubular
acoustic systems with varying cross section - the direct and inverse problems. Part 1:
15 D.B.Sharp