Non-destructive Testing of Tunnel Concrete Lining Using Wavelet

Transform

In-Mo Lee1, Il-Soo Cheon1, Joo-Gong Lee1, Gye-Chun Cho2

1
Department of Civil Engineering, Korea University, Seoul, Korea
2
Department of Civil and Environmental Eng., KAIST, Daejeon, Korea

ABSTRACT

The wavelet transform of transient signals provides a method for mapping the time domain into the
frequency spectrum as a function of time. To verify the availability of wavelet transform as a time-
frequency analysis tool, model experiments have been performed on the specially designed concrete
lining where defects (e.g., crack and cavity) are artificially made. The stress wave is generated by
dropping a steel ball and the acquired signal is analysed using wavelet transform as well as Fourier
transform. It is found that the contour map by wavelet transform renders more distinct results than
power spectrum by Fourier transform. The arrival time of each frequency component needed for
estimation of the wave velocity can be easily determined from ridge analysis. Furthermore, the
thickness of the concrete lining and the defect location can be effectively estimated using wavelet
transform. Therefore, it is concluded that the wavelet transform is an effective analysis tool for
detecting defect zones which exist in tunnel concrete linings.

1. INTRODUCTION

In the domestic construction industry in Korea, the demand of inspection and maintenance or repair
works for existing infrastructures has been increasing as well as that of new construction works. This
trend is also reflected on tunnel construction projects; about 40% of total budget for tunnelling is spent
on maintenance and repair works. The need of applying non-destructive testing (NDT) techniques for
assessing the soundness of the tunnel concrete lining has been increasing as well. To investigate the
safety and stability of the tunnel concrete lining, numerous studies have been conducted over the years
(Sansalone, 1997; Sansalone et. al., 1998; Yoon et. al., 2000). Several techniques have been developed
such as ground penetrating radar (GPR), ultrasonic method, spectral analysis of surface wave (SASW),
acoustic emission (AE) among others. However, these techniques in general require sophisticated
equipment and highly experienced operators to obtain meaningful results. This hinders the application
of NDT techniques to characterization of the soundness of tunnel concrete lining.
In this study, to effectively identify defects in tunnel concrete linings, a new NDT procedure is
developed using wavelet transform and is verified through experimental studies performed on a model
concrete lining.

2. SIGNAL PROCESSING

2.1 Limitation of Fourier Transform

Any waveform can be expressed as an infinite sum of sine waves and cosine waves and can be
expressed as follows.

B16 1
 ∧ ∞
f (ω) = ∫ f (t )e −iωt dt (1)
−∞
∧
where ω is the angular frequency and t is time. The Fourier coefficient, f (ω ) , is obtained by
correlating signal, f(t), with a sinusoidal wave, e-iωt. The Fourier transform provides insightful results
for most stationary signals. Meanwhile, in the analysis of a non-stationary signal, the Fourier
transform has some disadvantages. Small changes of a signal may not be realized and the analysis may
change depending on the length of data (Akin, 2002). Moreover, when a signal is converted into the
frequency domain by the Fourier transform, the information in the time domain has to be hidden. In
order to resolve such limitations in the Fourier transform analysis, several time-frequency analyses
have been attempted such as the analytical signal using instantaneous signal parameters, the short-time
Fourier transform (time windowing), the windowed Fourier transform (band-pass filters), and the
wavelet transform (Santamarina and Fratta, 1998).

2.2 Necessity of Time-Frequency Analysis

Applying an impact as a source on a medium generates a pulse with a wide frequency band whose
shape and characteristics change due to dispersion phenomenon. Complexities arise when the acquired
signals consist of the incident, refracted and reflected contents by defects and layers or boundaries.
This transient (or non-stationary) signal cannot be properly interpreted by Fourier transform analysis
which gives only information in frequency domain. However, time-frequency analysis is able to
analyse signals containing multiple propagation modes, which overlap and superimpose in the time
domain. One of the most powerful time-frequency analyses is wavelet transform analysis, which
provides how the frequency content of a given signal changes as a function of time. For illustration,
the comparison between wavelet transform and Fourier transform analysis for a stationary signal is
shown in Figure 1, while the comparison for a non-stationary signal is shown in Figure 2. In contrast
to the Fourier transform, the wavelet transform can distinguish the difference between the stationary
signal and the non-stationary signal. Details on wavelet transform analysis used in this study are
discussed next.

(a) Frequency domain analysis (b) Time-frequency analysis

Figure 1 Comparison between Fourier transform and time-frequency analysis for a stationary signal

(a) Frequency domain analysis (b) Time-frequency analysis

Figure 2 Comparison between Fourier transform and time-frequency analysis for a non-stationary
signal

2.3 Wavelet Transform

B16 2
The concept of wavelet transform as a specialized field can be traced back to the work done by
Grossman and Morlet (1984). The wavelet transform has been recognized to overcome some
limitations of Fourier transform and has been widely used in various fields such as quantum
mechanics, speech recognition, image processing and signal processing, and geophysical process
(Newland, 1994). A wavelet is considered as a window function both in the time and frequency
domain, allowing a time-frequency analysis of a signal. Thus, the wavelet transform can achieve good
time localization, which is appropriate to estimate the arrival time of wave components travelling at
different velocities (Daubechies, 1990).
To analyse signal structures of very different sizes, it is necessary to use time-frequency atoms with
different time supports. The wavelet transform decomposes signals into dilated and translated
wavelets. Consider a real or complex continuous-time function, ψ(t), with the following properties:
∞
∫−∞
ψ (t )dt ≈ 0 (2)
and
∞
∫
2
ψ (t ) dt < ∞ (3)
−∞
The function, ψ(t), is a mother wavelet or wavelet if it satisfies these two properties. Mathematically,
the wavelet transform of a signal is the cross-correlation of the signal with the wavelet, for wavelets of
varying central frequency. Thus, its physical meaning is to identify similarities between the signal and
the wavelet kernel that represents the wavelet as it is time shifted along the signal. The wavelet
transform is defined as a summation of a signal multiplied by the dilated (or contracted) and translated
wavelet in continuous time:
+∞ 1 * t −u
Wf (u , s ) = 〈 f ,ψ u ,s 〉 = ∫ f (t ) ψ ( )dt (4)
−∞
s s
and
1 t −u
ψ u ,s (t ) = ψ( ) (5)
s s
where f(t) is the original signal, ψ(t) is the mother wavelet, u is the translating parameter (i.e., the
wavelet is time-shifted according to u), s is the scaling (e.g., dilating or contracting) parameter (i.e.,
the wavelet is time-scaled as indicated by s), and the asterisk ∗ denotes the complex conjugate. The
normalizing factor 1 / s ensures energy conservation during the wavelet transform. Since wavelet
has a zero average, the wavelet integral measures the variations of f in a neighbourhood u, whose size
is proportional to s. When the scale s goes to zero, the decay of the wavelet coefficients characterizes
the regularity of f in the neighbourhood u. This has the important applications for detecting transients
and analyzing fractals.
A local time-frequency energy density, Pωf, which measures the energy of a signal f in the Heisenberg
box of each wavelet ψu,s(t) centered at (u, ξ=η/s), is defined as follows:
2
⎛ η⎞
2
Pω f (u , ξ ) = Wf (u , s ) = Wf ⎜⎜ u , ⎟⎟ (6)
⎝ ξ⎠
Therefore, the squared magnitude of the wavelet transform using Gabor wavelet corresponds to the
energy spectrum, called a scalogram. The normalized scalogram is defined as follows (Mallat, 1998):
2
ξ Wf (u , s )
Pω f (u , ξ) = (7)
η s
The normalized scalogram measures the energy of a signal f(t) in a time-frequency plane. The ridge
algorithm computes instantaneous frequencies from the local maxima of scalogram. The results of
ridge analysis for a non-stationary signal give useful information about the energy arrival time of each
frequency.

B16 3
3. EXPERIMENTAL STUDY

3.1 Construction of Model Tunnel Concrete Lining

A model concrete lining is constructed to simulate in-situ tunnel concrete linings as closely as
possible. The model concrete lining is designed reflecting the summary report of safety diagnosis on
21 tunnels in Korea which have 166 locations of cracks and 177 locations of cavities (Korean Ministry
of Construction and Transportation, 2002). The model concrete lining is made of a plain concrete with
a compressive strength of 23.5 MPa at 28 days curing, and has a dimension of 4m in width, 4m in
length, and 3m in height. When designing the model concrete lining, a sound part, a thinner part, a
cavity part, and a crack part are placed based upon the general trends documented in the report. The
cavity is shaped by placing a small box and the crack is generated by placing and removing a steel
plate in the model concrete lining during curing.

3.2 Experimental Setup

The measurement system consists of transducers, a source and a digitizer. The transducers and sources
are located in a line to monitor waves propagating from the source to the first transducer and then to
the second transducer. Each broad band transducer (e.g., accelerometer) attached at the surface of
concrete lining is connected to the digitizer. During the entire test, the input signal to the second
transducer is also recorded simultaneously when the first transducer is triggered at 0.01sec. A transient
stress pulse is artificially generated by dropping a steel ball on the surface of the concrete lining. A
signal is digitized into 5,000 samples at a sampling rate of 100,000 samples/sec. The digitized signal is
then stored in the computer for the post-process. The schematic diagram of the experimental setup is
shown in Figure 3.

3.3 Data Reduction

Surface wave velocity. The surface wave velocity can be estimated as follows:
L
VR = (8)
∆t
where L is the distance and ∆t is the time difference between the two transducers.
Thickness of the concrete lining. The thickness of the concrete lining can be estimated by defining a
non-dispersive frequency. The wavelength of the surface wave is closely related to the exploration
depth. Therefore, the thickness of tunnel concrete lining can be simply estimated by using the
following relationship between the wave velocity (VR) and the minimum frequency (fr) appeared in the
non-dispersive zone:
V
λ= R (9)
fr
Also the depth (D) represented by surface wave is considered as one half of the wavelength
(Heukelom and Foster, 1960). Therefore, the thickness of the model concrete lining can be evaluated
as follows:
λ (10)
D=
2
Location of crack. The location of crack can be measured by obtaining the surface wave velocity. By
using the result of ridge analysis on the signal obtained at the first transducer, the time difference (∆t)
between the incident wave and the reflected wave can be directly evaluated in the time-frequency
domain. Therefore, the distance (L) between the transducer and the crack location can be determined
from the relationship between the surface wave velocity (VR) and the time difference (∆t) as follows:
L = VR ⋅ ∆t (11)

B16 4
 Figure 3 Experimental setup Figure 4 Sound concrete lining model

4. TEST RESULTS – DISCUSSION

4.1 Results at a Sound Concrete Lining

Measurements are conducted on the sound concrete lining model as shown in Figure 4. Two
transducers are installed on the concrete lining surface to determine the surface wave velocity. The
location of the transducers and a source is displayed in Figure 4 as well.
Wavelet transform results obtained from the acquired signals are presented in Figures 5, 6 and 7. The
contour map obtained by wavelet transform shows that a high intensity peak is located at around 11
kHz (i.e., the energy of the generated stress wave is concentrated at around 11 kHz).

(a) The first transducer (b) The second transducer

Figure 5 Ridge analysis for the signal received at the two transducers

Figure 6 Results of ridge analysis Figure 7 Time delay

B16 5
As shown in Figure 7, the arrival time difference is estimated from the results of ridge analysis in the
frequency range between 4 and 30 kHz where the surface wave velocity is almost constant. By using
the time difference (i.e, 0.00026 sec) and the distance between two transducers (i.e., 0.6 m), the
surface wave velocity, VR2, can be estimated regardless of variation of frequencies as follows:
0 .6 m
VR 2 = ≈ 2310 m / sec
0.00026 sec
The lower limit (i.e., 4 kHz) of the frequency range which shows non-dispersion phenomenon in
Figure 7 is converted to wavelength (λ) using the surface wave velocity (i.e., 2310 m/sec) (refer to
Equation 9). The estimated wavelength is 57.7 cm. Therefore, the thickness of the model concrete
lining can be calculated using Equation (10) as follows:
57.7
D= ≅ 28.9 cm
2
The calculated value is close to the thickness of the model concrete lining (i.e., 30 cm).

4.2 Results at Crack Zone

The object of this experiment is to detect the location of surface-opening cracks that can occur in the
tunnel concrete lining. The schematic drawing of experimental setup on the concrete lining with a
crack is shown in Figure 8. The test results in time domain are analysed using wavelet transform and
the ridge analysis is performed. The time delay (i.e., arrival time difference) between the incident and
the reflected waves is shown in Figure 9.

Figure 8 Concrete lining with a crack Figure 9 Time delay between the incident wave
and the reflected wave

The location of the defect (i.e., the crack) can be determined from the obtained time delay. The
averaged time difference over the frequency range between 4 and 50 kHz is 0.000285 sec. Therefore,
the distance (L) from the first transducer to the crack can be calculated from the surface wave velocity
(i.e., 2310 m/sec) and the averaged time difference (0.000285 sec) as follows:
2310 m / sec × 0.000285 sec
L= = 0.328m
2
The calculated distance (i.e., 32.8 cm) is close to the real distance (i.e., 30 cm).

4.3 Results at Cavity Zone

Measurements are made on a model concrete lining with a cavity in order to investigate the effect of
cavity presence. The schematic drawing of the model concrete lining with a cavity is shown in Figure
10. The first transducer is located in front of the cavity and the second transducer is located behind the
cavity. The distance between the first and second transducers is 120 cm and the distance from the
source to the first transducer is set to be 30 cm in order to avoid the near field effect.

B16 6
The tests results are shown in Figures 11 and 12. By comparing those results, it is obvious that the
change of the dominant frequency occurs when a signal is passing through a cavity. That is, while the
dominant frequency of the incident wave received in front of the cavity is 30.5 kHz, the dominant
frequency of the modulated incident wave received behind the cavity decreases up to 10 kHz.
Therefore, it is concluded that the generated wave may be modulated and the dominant frequency may
be altered when the signal passes through a cavity region. For illustration, the specific frequency (or
scale) components (i.e., 10 kHz and 30.5 kHz) are extracted from time-frequency domain and the
distributions of wavelet coefficients for 10 kHz and for 30.5 kHz are shown in Figure 13 and Figure
14, respectively. It is observed that the presence of a cavity acts as a low-pass filter.

Figure 10 Concrete lining with a cavity

(a) Signal displayed in the time domain (b) The time-frequency analysis
Figure 11 Signal measured at the first transducer with a cavity

(a) Signal displayed in the time domain (b) The time-frequency analysis
Figure 12 Signal measured at the second transducer with a cavity

(a) First transducer signal (b) Second transducer signal

Figure 13 Distribution of wavelet coefficients at frequency (= 30.5 kHz)

(a) First transducer signal (b) Second transducer signal

Figure 14 Distribution of wavelet coefficients at frequency (= 10 kHz)

B16 7
5. CONCLUSION

A new analysis method for non-destructive testing was developed to effectively investigate defects
existed in the tunnel concrete lining using surface waves. A comparative analysis between Fourier
transform and wavelet transform suggests that the wavelet transform should be considered as an
appropriate analysis tool for a transient or non-stationary signal. Experimental studies performed on
the model tunnel concrete lining show that the surface wave analysis by using wavelet transform
makes it easy to estimate surface wave velocity, to determine the concrete lining thickness, to detect
the crack location, and to characterize the presence of a cavity. Over all, it is clear that the wavelet-
based analysis is promising for assessment of the soundness of tunnel concrete linings.

REFERENCES

Akin, M., 2002. Comparison of wavelet transform and FFT methods in the analysis of EEG signals.
Journal of Medical Systems, Vol. 26, No. 3, pp241~247.
Daubechies, I., 1990. The wavelet transform, time-frequency localization and signal analysis. IEEE
Transactions on Information Theory, Vol. 36, No. 5, pp 961~1005.
Grossman, A. and Morlet, J., 1984. Decomposition of hardy functions into square integrable wavelets
of constant shape. SIAM J. of Math. Anal., 15(4), pp723~736.
Heukelom, W and Foster, C. R., 1960. Dynamic testing of pavements. Journal of the Structural
Divisions, ASCE, SM1, pp. 1~28.
Korean Ministry of Construction and Transportation, 2002. A development of non-destructive testing
technique using sound wave analysis. Final Report, 41-01
Mallat, S., 1999. A wavelet tour of signal processing. Academic Press, Second Edition.
Newland, D. E., 1994. Wavelet analysis of vibration, Part 1: Theory. Journal of Vibration and
Acoustics, Vol. 116, pp409~416.
Newland, D. E., 1994. Wavelet analysis of vibration, Part 2:Wavelet maps. Journal of Vibration and
Acoustics, Vol. 116, pp417~425.
Prosser, W. H. and Seale, M. D., 1999. Time-frequency analysis of the dispersion of lamb modes.
Journal of Acoustics American, Vol. 105, No. 5, pp 2669~2676.
Sansalone, M., 1997. Impact-echo: The complete theory. ACI Structural Journal, Vol. 94, No. 6, Title
no. 94-S71.
Sansalone, M., Lin, J. M. and Streett, W. B., 1998. Determining the depth of surface opening cracks
using impact-generated stress waves and time-of-flight techniques. ACI Materials Journal, Vol. 95, pp
159~177.
Santamarina, J.C., and Fratta, D., 1998. Introduction to discrete signals and inverse problems in civil
engineering. ASCE Press, p. 327.
Yoon, D. J., Weiss, W. J. and Shah, S. P., 2000. Assessing damage in corroded reinforced concrete
using acoustic emission. Journal of Engineering Mathematics, Vol. 126, No. 3, pp273~283.

B16 8