Construction and Building Materials 194 (2019) 42–50

Contents lists available at ScienceDirect

Construction and Building Materials

journal homepage: www.elsevier.com/locate/conbuildmat

Embedding ultrasonic transducers in concrete: A lifelong monitoring

technology
Arnaud Deraemaeker ⇑, Cédric Dumoulin ⇑⇑
Université libre de Bruxelles (ULB), École polytechnique de Bruxelles, Building, Architecture and Town Planning department (BATir), Belgium

a r t i c l e i n f o a b s t r a c t

Article history: This paper deals with lifelong monitoring of concrete structures using embedded piezoelectric transduc-
Received 24 April 2018 ers. Thanks to such transducers, monitoring of the concrete can be automated both at the early age, right
Received in revised form 22 October 2018 after the concrete is cast, and over the long term, until the end of the lifetime of the structure. For long
Accepted 1 November 2018
term monitoring, the wave velocity variations introduced by operational factors effects should be filtered
Available online 8 November 2018
out. In the present study, we introduce a new technique called the Direct Wave Interferometry (DWI)
which uses time stretching only on the early part of the recorded wave. This appears to be a good tradeoff
Keywords:
between the high resolution of the Coda Wave Interferometry (CWI) for low velocity variations and the
Ultrasonic wave
Embedded piezoelectric transducer
reliability of the measurement of the time of flight (TOF) for large velocity variations.
Direct wave interferometry (DWI) Ó 2018 Elsevier Ltd. All rights reserved.
Concrete assessment
In situ monitoring

1. Introduction systems which allow to monitor remotely and wirelessly the evo-
lution of the compressive strength. There is a clear demand from
Concrete is the most widely used construction material in the the industry for such automated solutions. The method does not
world. It is a complex material whose mechanical properties allow however to estimate the Young’s modulus of concrete, which
evolve continuously from the time of casting until the end of its can be of interest at early-age for structures with cantilevers.
lifetime. During the construction phase, the main issue for the Over the time, when the structure is in service, the mechanical
operator is to determine the time when the formworks can be properties of concrete can also deteriorate due to several mecha-
removed which depends mainly on the evolution of the compres- nisms such as the loss of prestress in prestressed concrete or chem-
sive strength. The time when post-tensioning can be applied is also ical attacks such as alkali-silica reactions or delayed ettringite
governed by the evolution of the compressive strength. Nowadays, formation. In order to ensure the safety of the infrastructure, it is
such evolution is assessed through very conservative norms or important to be able to assess its remaining strength. Destructive
compressive tests on small specimen which have been cast at the compressive tests can be performed on sampled cores extracted
same time as the structure and stored at the same location. There from the structure. For obvious reasons however, non-destructive
is a certain degree of inaccuracy associated to this technique as the techniques are preferable. Among these, the ultrasonic pulse veloc-
temperature evolution in the actual structure is not the same as ity method (UPV) is currently one of the most widely used tech-
the one in small specimens, especially when the structure is mas- niques. It is an established technique described in different
sive. In order to overcome this drawback, maturity methods [1,2] national norms among which the European standard EN 12504-4
which are based on the measurement of the actual temperature [3] or the ASTM norm C 597-02 [4], and used in many commercial
evolution inside the concrete, calibrated laboratory compressive systems. It consists in exciting the emitter by a short pulse signal
tests on small specimen and the knowledge of the activation and measuring the time of propagation of the wave from the emit-
energy can be used to get a more accurate estimation of the com- ter to the receiver. Knowing the distance between the transducers,
pressive strength. There exist currently automated maturity the velocity of the fastest wave (the compressive wave, also called
the P-wave) can be measured. The P-wave velocity is related to the
mechanical properties (Young’s modulus and Poisson’s ratio) of the
⇑ Principal Corresponding Author. concrete and can therefore be used to assess the uniformity and
⇑⇑ Corresponding author.
relative quality of concrete [5]. It can also indicate the presence
E-mail addresses: [email protected] (A. Deraemaeker), [email protected]
of voids and cracks [6] and can be used to evaluate the setting time
(C. Dumoulin).
URL: http://batir.ulb.ac.be (A. Deraemaeker). in fresh concrete [7]. Provided an assumption is made on the value

https://doi.org/10.1016/j.conbuildmat.2018.11.013
0950-0618/Ó 2018 Elsevier Ltd. All rights reserved.
 A. Deraemaeker, C. Dumoulin / Construction and Building Materials 194 (2019) 42–50 43

of the Poisson’s ratio, it can be used to give an estimate of the in service, the temperature, the moisture or the stress state have
Young’s modulus. The wave velocity has also been shown to be an impact on the measured ultrasonic signals which needs to be
correlated to the concrete compressive strength, but as this rela- distinguished from the impact of the evolution of the microstruc-
tionship depends on many parameters of the concrete, a calibra- ture (damage).
tion curve needs to be established for a specific concrete [8]. This paper focuses on two important aspects of real-time mon-
The typical frequency range for UPV testing is from 20 kHz to itoring: monitoring of the hardening phase of concrete structures,
200 kHz. In that frequency range, wave propagation in concrete and long-term damage monitoring. For both aspects, the same
is a complex phenomenon due to the high heterogeneity of the monitoring hardware is used, which consists in pairs of embedded
microstructure: the wave sent from the emitter will interact with ultrasonic transducers where the emitter is excited via a pulse
the aggregates causing multiple reflections. The initial wave will excitation, and the signals are recorded at the receiver side. The
be split in many different components traveling different paths, post-processing is however different for each application: for the
and reaching the receiver at different times, resulting in a very long hardening phase, automated UPV measurements are performed
recorded signal. Typically, for a pulse input of five microseconds, in order to follow the evolution of the Young’s modulus at early
the signal recorded at the receiver side will be several milliseconds age. For the damage monitoring, a novel method is proposed for
long. Although there is no strict limit, one can distinguish between the filtering of the environmental effects. It is inspired from the
the early-wave, also called the ballistic wave, and the later wave, approach used in CWI based on time stretching and the definition
called the CODA. The wave components which reach the receiver of a correlation coefficient, but is applied to the early part of the
first have travelled the shortest path, and therefore interacted wave instead of the CODA. The new method, called direct wave
mainly with the microstructure in the direct line of sight between interferrometry (DWI) is shown to be better suited for on-line
the emitter and the receiver. The UPV is therefore a measure of the damage monitoring under changing operational conditions. This
properties of concrete between the emitter and the receiver, while is because the time stretching method is efficient to remove oper-
the late arrival wave carries information about more distant loca- ational variability only when applied to signals which have trav-
tions, but is much more complex to interpret. While still at the elled through areas in which the stress state, the temperature
stage of research, there have been a few attempts at extracting and the moisture are more or less uniform, which is generally
information from the late CODA. The most widely used technique not the case when later parts of the wave are considered. The
is coda wave interferometry (CWI) and has been used to estimate use of the early wave also preserves the local nature of the damage
the acoustoelastic parameters of concrete in [9,10]. Providing a indicator.
dense network of sensors is used, it is also possible to locate dam- In this paper, the efficiency of the proposed monitoring system
age in a concrete structure using more advanced techniques based based on embedded piezoelectric transducers is demonstrated on
on coda wave interferometry [11–13]. two small concrete beams. The test specimens, the embedded
There are several drawbacks to the techniques described above. transducers and the testing hardware are presented in Section 2.
None of them is suited to monitor concrete over the whole lifetime, The monitoring of the hardening phase and the damage monitoring
from the time of casting. Maturity methods are limited to the early are presented in Sections 3 and 4. The hardening phase is moni-
age (first few days) and cannot give information about the deteri- tored by computing the evolution of the UPV based on the estima-
oration of concrete over time or the early cracking. Ultrasonic test- tion of the time of flight, and deducing the evolution of the Young’s
ing, including UPV and CWI using external probes cannot be used modulus. One of the samples is then loaded under a three-point
on site at early age due to the presence of formworks, requires bending test and damage monitoring is performed. The operational
manpower and is difficult to apply over the long term due to acces- variability is due to the evolution of the stress state in the beam
sibility issues. Over the last ten years, several researchers have which, due to the acoustoelastic effect (change of wave velocity
studied the possibility of embedding the ultrasonic probes directly due to the applied stress), impacts the wave velocity even in the
inside the concrete. The first and obvious advantage of the tech- absence of damage. The novel approach based on DWI is applied
nique is the potential automation of ultrasonic monitoring. Addi- to filter out successfully the acoustoelastic effect. It is also shown
tional advantages are an enhanced coupling (without coupling that applying time stretching before computing the damage indica-
agents) with the concrete for stronger waves generation over tor is equivalent to tracking the change of amplitude of the first
longer paths, the protection of the transducers from environmental peak of the recorded waves [30].
and accidental attacks, the possibility to perform measurements in
the presence of formworks, and an added flexibility of transducers
arrangements which allows for direct wave propagation paths. A 2. Specimens and monitoring system
major requirement with such transducers is however their low-
cost as they will be lost in the structure. Experiments are performed on two short non-reinforced con-
Such kinds of transducers have been successfully used for the crete beams in which a pair of transducers has been embedded
monitoring of (i) cement and concrete hydration [14,15], (ii) the with a spacing of 14.3 cm (Fig. 4). The beams are made of ordinary
evolution of the compressive strength of concrete at early age concrete. The geometry and the components of the beams are
[16,17], (iii) concrete cracking [18–22], (iv) water seepage given in Table 1.
[23–25] and (v) mechanical properties of concrete [26] including The embedded transducers have been developed at ULB-BATir
the acoustoelastic effect in compression [27]. At ULB-BATir, we and are made of a thin Lead Zirconate Titanate (PZT) patch which
have worked on the development of our own embedded transduc- is a piezoelectric material [31,32] and surrounded by several coat-
ers to measure the P-wave velocity at early-age [28] and monitor ing materials (Fig. 1). The use of such embedded transducers
cracking in several laboratory tests [29,30] (pull-out, three-points allows for direct measurement instead of indirect, semi-direct or
bending and compressive tests). through the thickness measurements [31,33].
Although limited to laboratory or short-time monitoring, all of The ultrasonic monitoring system is described in Fig. 2. It is
these applications show the high potential of ultrasonic testing composed of a data acquisition board NI PXI-6115 (10 MHz sam-
with embedded transducers to perform real-time on-line monitor- pling rate) and a variable gain pre-amplifier SmartPre (+18 to
ing of concrete infrastructure. An important step forward is how-
60 dB, band-pass 1 kHz–1 MHz) designed by SMARTMOTE which
ever needed, as long term monitoring involves environmental are controlled by an in-house program based on the LabView pro-
variability which needs to be filtered out. It is known that when gramming environment. The system is able to perform up to 150
44 A. Deraemaeker, C. Dumoulin / Construction and Building Materials 194 (2019) 42–50

Table 1 systems which generally involve high voltage (400–800 V) pulsers

Composition of the concrete and geometry of the Non-Reinforced Concrete (Mini) to increase the signal to noise ratio (SNR) and hence the accuracy of
Beams
the evaluation of the onset time. This is particularly important for
Geometry of the Non-Reinforced Beam the monitoring of the hardening phase due to the very low ampli-
Llh cm 40  10  10 tude of the recorded signals in fresh concrete. Our approach to
Distance between transd. cm 14:3
increase the SNR for early age monitoring is to benefit from the
Concrete components Density high measurement rate of our system (which cannot be reached
CEM I 52.5 N PMES CP2 (Saint Vigor, FR) kg=m
3 340
with high-voltage pulsers) and compute the average of one hun-
Sand 0=4 (Bernières, FR) kg=m
3 739
dred successive measurements. In Section 3, one signal consisting
Gravel 8=22 (Bernières, FR) kg=m
3 1072
of an average of 100 signals measured at a rate of 25 Meas./s is
Total water kg=m
3 184
recorded every four minutes in order to follow the evolution of
Total wet density kg=m
3 2335
the P-wave velocity during the first 70 h after casting the concrete.
Total dry density kg=m
3 2250
For the bending test, the initial SNR is much better than in fresh
Mechanical Properties concrete: the wave is much less attenuated in hard concrete, so
Young’s Modulus⁄ Ecm GPa 36
that an average of only ten signals is used. The monitoring is per-
Compressive Strength (cube)⁄ f ck;cube MPa 38:04
formed continuously at a rate of 25 Meas./s. Due to the very high
Compressive Strength (cylinder)⁄ f ck MPa 30:05
Tensile Strengthy f ctm MPa 2=3 measurement rate in this test and the occurrence of damage, some
0:3f ¼ 2:89
 ck  recorded signals are polluted with acoustic emission events which
Flexural Strengthy f ctm;fl MPa hðmÞ
1:6
1000 f ctm ¼ 4:63
have to be filtered out. Acoustic events are waves generated by the
Expected Maximum Loady kN 8:1
sudden release of energy due to cracking events [34,35]. In the pre-
y
Estimated from EUROCODE EN 1992-1-1:2004. sent study, it was decided to remove these signals from the data.
⁄
Measured. This is not problematic as there are enough non-polluted signals
due to the very high measurement rate. Another option would
have been to increase the voltage at the emitter side, but this is
not the choice made in this study, as explained above. The mea-
surements which are presented in Section 4.3 correspond to an
average of the last 10 sane signals (i.e. without acoustic emission
events). The procedure which is used to detect and remove the pol-
luted signals consists in a cross correlation test with the average of
the previous ten non polluted signals:
R tf
0 Sj ðt ÞSm ðt Þdt
AEj ¼ qRffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
R tf 2 ð1Þ
tf 2
0 Sj ðt Þdt 0 Sm ðt Þdt

Fig. 1. Embedded piezoelectric transducers designed and manufactured at ULB-

th
BATir. where Sj ðt Þ and Sm ðt Þ are respectively the j signal and the average
of the last ten sane signals. t f is the measurement time (here 3ms).
This indicator is theoretically equal to 1 for two identical signals. As
the measurement rate is much higher than the rate of change due to
damage, it can be considered that then successive signals should be
almost identical. When an acoustic event occurs, the local cross cor-
relation coefficient drops strongly due to the drastic change of
shape of the signal, which allows for discriminating those signals
from the sane signals [30]. In practice, because of the inevitable
noise on the recorded signals, and the slow evolution of the signal
with damage, the indicator is not strictly equal to 1 for a signal
without acoustic event so that a threshold has to be considered.
We have found that a value of threshold of 0:9 allows for removing
all the signals polluted with acoustic emission events. The first ten
sane signals are taken when no load is applied to the specimen so
that there are no acoustic emission events. The value of AEj is then
computed for each new measurement and if it is sane (AEj > 0:9), it
is included in the stack of the last ten sane signals (see Fig. 3) for
Fig. 2. Fast (6 150 ultrasonic measurement per second) and low-voltage (6 10 V) averaging.
data acquisition system. The DAQ is a NI PXI-6115 (out 4 MS/s, in 10 MS/s) and the
pre-amplifier is a SMARTMOTE SmartPre. The DAQ and the pre-amplifier (gain) are
automatically controlled by a LabView based in-house program.

ultrasonic measurements per second while the signal which is gen-

erated is a short pulse of 5 ls corresponding to a frequency band
up to 200 kHz (
6 dB at 120 kHz). For the present experiments,
the ultrasonic measurement rate has been set to 25 Meas./s. The
duration of each measurement is 3 ms which allows for a complete
attenuation of the ultrasonic wave before the next measurement.
Fig. 3. Process to remove the signals polluted by undesired events such as acoustic
The excitation signal is a low voltage short pulse (10 V, 5 ls). emission events: each signal is compared with the average of the last 10 sane
Such a voltage is very low in comparison to commercial ultrasonic signals.
 A. Deraemaeker, C. Dumoulin / Construction and Building Materials 194 (2019) 42–50 45

Fig. 6. Evolution of the P-Wave Velocity as a function of the time since casting.

Z 1  
1 t
s
C ðk; sÞ ¼ pffiffiffi xðt ÞW dt ð3Þ
k
1 k
where the wavelet W is a Complex Gaussian Wavelet.
The different devices of the monitoring system and the layers
surrounding the transducers necessarily result in a certain time
delay in the sensor-actuator line so that the system should be cal-
ibrated for accurately determining the time-of-flight between the
sensor and the actuator. This is performed by measuring the travel
time in water [28], where the wave velocity is accurately known.
Knowing the distance between the emitter and the receiver and
the delay, the velocity of the fastest wave (the P-wave) can be
deduced. Fig. 6 shows the evolution of the P-Wave velocity during
the first 70 h after casting the concrete. It can be observed that the
typical S-Curve which describes the hydration kinetics is well
Fig. 4. Specimens used in the study. The concrete property and the geometry of the
beams are given in Table 1.
caught by the monitoring system from very early age.
The dynamic E-modulus can therefore be deduced from the P-
Wave velocity [38]:
  
3. Monitoring of the hardening phase using UPV measurements 1 þ mdyn 1
2mdyn
Edyn ¼ qV 2P   ð4Þ
1
mdyn
The P-Wave velocity is computed from the known distance
between the emitter and receiver and the estimated time of flight Assuming a value of the dynamic Poisson’s ratio of mdyn  0:3 at
of the ultrasonic wave. The onset time of the ultrasonic wave can 28 days [38], the dynamic E-modulus is 36 GPa.
be automatically estimated by finding the minimum of the Akaike
Information Criterion (AIC) computed in a given interval (Fig. 5). 4. Long term damage monitoring
The AIC is computed as follows:
A three-points bending test is performed on one of the two non-
AICðkÞ ¼ k log ðVarðS½1; kÞÞ þ ðN
k
1Þ log ðVarðx½k þ 1; N ÞÞ
reinforced concrete beams after 28 days. An overview of the test
ð2Þ setup is shown on Fig. 7. The loading machine is a 200 kN hydraulic
th jack bending testing machine. The force and the displacement have
where Varð X Þ denotes the variance, k is the k sample of the signal
been recorded on a computer using a National Instruments DAQ
SðkÞ of length N. This time window is defined around a first estimate system (NI PXI-4461). The vertical displacement at the center of
of the time of arrival found as a threshold of the amplitude of the the beam is measured by an inductive displacement transducer
envelope of the signal [36,32]. The envelope function is calculated (5mm LVDT, HBM W5TK) wired to the DAQ through a specific sig-
from the wavelet transform C ðk; sÞ [37] of the signal at a low scale nal conditioner. The ultrasonic monitoring system is described on
(i.e. low frequency) k Fig. 2. During the bending test, short low voltage pulses are sent
at regular interval (5 ls, 10 V, 25 Signals/s).

4.1. Loading procedure

Since the beam is not reinforced, the maximum load which can
be applied is reached at the initiation of the crack. The loading
machine is controlled in displacement thanks to an inductive linear
displacement transducer (LVDT) as shown in Fig. 7. In order to make
sure that the crack is initiating at the center of the beam, a notch of
1 cm of depth is sawed (8). The loading procedure and the force-
displacement curve are displayed on Fig. 8 where different pictures
Fig. 5. Estimation of the onset time with the AIC. allow to track the crack opening. The beam is unloaded right after
46 A. Deraemaeker, C. Dumoulin / Construction and Building Materials 194 (2019) 42–50

Fig. 7. Overview of the test setup for the bending test. The geometry of the specimen is given in Table 1. The loading machine is a 200 kN hydraulic jack bending testing
machine.

Fig. 8. Loading Procedure and Force-Displacement curve related to the visual appearance of the crack. The three point bending test is controlled in displacement using a
LVDT.

the crack visually appears. The different pictures are numbered dV dtr t r
t r;0
¼ 
¼
ð5Þ
(from 1 to 3) in order to relate them to. (a) the force- V0 tr;0 tr;0
displacement and force-time curves on Fig. 8, (b) the relative
where t r and tr;0 are respectively the onset time of the current
velocity variation displayed on Fig. 10, and (c) the evolution of
signal Sðt Þ and the onset time of the baseline signal S0 ðt Þ. It is
the damage index on Fig. 13. The crack visually appears when the
known that the direct evaluation of the time of flight leads to
maximum load is already exceeded (see 2 in Fig. 8). It is important
high uncertainties as it clearly appears in the experiments
to note that the sudden load decreases which can be observed in the
presented hereafter (Fig. 10). The precision of  depends
loading phase are due to technical features of the loading machine.
directly on the sampling rate of the measurement (10 MHz
in the present case).
4.2. UPV measurements (II) The second method is based on the classical coda wave inter-
ferometry (CWI) method. This method is known to be very
When in service, a concrete structure is subject to operational precise to track small velocity variations due to the temper-
variability due to changing loads (traffic, wind, . . .) and environ- ature effects or small stress variations (acoustoelastic effect).
mental effects (temperature, humidity). It is known that the wave The total wave field is simply the summation of all the waves
velocity varies to a certain extent due to these changing conditions, that propagate along all the possible paths. One can therefore
as well as due to the damage. These relative velocity changes are describe a multiply scattered wave SðtÞ as the superposition
very small compared to the changes of velocities at the early age. of the different components of an initial wave packet with
In this paper, we explore and compare different methods to esti- random amplitudes and delays. Consequently, the effect of
mate in real time the relative wave velocity variation using embed- a global velocity change  ¼ dV=V in the medium is to stretch
ded piezoelectric transducers with the focus on the efficiency to the original signal S0 ðt Þ. The stretching factor  is determined
monitor very small changes of velocities: by searching for the maximum cross-correlation coefficient
CC ð) in all the duration of the CODA (see Fig. 10).
(I) The first method consists in estimating the onset time of the R tf
received wave (UPV) as described in Section 3. Considering a SðtÞS0 ðt ð1 þ ÞÞdt
CCðÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
t0
small variation of the velocity, the relative velocity change is R tf 2 Rt ð6Þ
t0
S ðt Þdt t0f S20 ðtð1 þ ÞÞdt
therefore given by
 A. Deraemaeker, C. Dumoulin / Construction and Building Materials 194 (2019) 42–50 47

where t 0 and t f depend on the time window in which the

cross correlation is estimated. The strategy which is adopted
is to use the MATLAB optimization function fminbnd to find
the minimum value of the decorrelation factor
(DC ¼ 1
CC). When the maximum correlation is high, 
can directly be used as an estimate of the relative wave veloc-
ity change. This will only be the case if the velocity change is
uniform in a large area around the transducers. The extent of
this area depends mainly on the time of arrival of the latest
waves which have travelled a much longer path and poten-
tially reached much more distant locations than the area
around the transducers. In most practical applications, the
stress field and environmental parameters are not uniform
in such a large area, so that the method is not applicable.
(III) As an alternative, the third method, which is introduced in
the present study, is directly inspired by the CWI, but it is
applied on the direct wave instead of the CODA (see
Fig. 9). The introduction of the direct wave interferometry
(DWI) is motivated by the fact that this part of the wave is
mainly impacted by variations in a restricted area located
along the line of sight from the source to the receiver
[39–41]. In such a restricted area, it is expected that the
wave velocity changes will be almost uniform, making the
measurement robust to complex loading conditions and
effects of boundary conditions. The use of a shorter wave
will result in a lower resolution than CWI because the delays
in the early part of the signal are smaller than the delays in
the CODA.

The initial (t i ) and final times (t f ) used for the computation of

Fig. 10. Evolution of the relative velocity change as function of the time (bending
the correlation coefficient CC for CWI and DWI are shown in test). The ’time of flight’ line is obtained by denoising (wden in MATLAB) the actual
Fig. 9. For CWI, we consider that the wave path should be longer measurements (green dots). (For interpretation of the references to colour in this
than several (at least 2 or 3) times the elastic mean free path ls figure legend, the reader is referred to the web version of this article.)
(the distance between successive scattering events, see [41]). For
concrete the elastic mean free path ls is usually around 15 to (a) the direct estimation of the relative velocity variation from
20cm at 200kHz [42,12]. For a wave velocity of 4500 m=s; 3ls corre- the TOF leads to very noisy results. In particular, one can
sponds to a time of flight of 133ls. For DWI, tf is based on the observe that the resolution is constrained by the sampling
observation that before that instant, signals are not changing rate of the measured wave (see the individual dots corre-
before cracking occurs. At that time, the wave has only encoun- sponding to the actual measurements in Fig. 10). Before
tered a few scattering events. Further investigations are needed denoising, the accuracy of the relative wave velocity change
in order to link this value to an intrinsic measure which should is in the order of 2% (2:10
2 ). After denoising, the resolution
be related to the stress gradients and relative positions of bound-
is improved and reaches about 2:10
3 .
ary conditions, but this is outside of the scope of the present study.
The evolution of the relative wave velocity change during the (b) the resolution of CWI is about 2:10
5 , as can be estimated
three-point bending test is shown in Fig. 10 using the three meth- from Fig. 10 (bottom). The later result is in accordance with
ods. The line corresponding to the direct measurement of the TOF the findings in [9]. Fig. 11 shows however that for CWI, the
is obtained by denoising (wden in MATLAB) the values actually wave starts to be decorrelated from the baseline after a time
found. One can observe that. of approximately 3 min. When the signal is decorrelated,
time stretching will fail to give a physical value of the rela-
tive wave velocity variation. CWI clearly fails to catch the
strong velocity variations due to damage after 10 min while
these are obvious from the TOF measurements.
(c) the resolution of DWI is about 2:10
4 . The decorrelation
starts to increase around 11 min which corresponds to the
time of the initiation of the crack. The technique can there-
fore be used with confidence before damage appears to track
small relative wave velocity changes, with an accuracy
which is a compromise between TOF and CWI. When dam-
age occurs, the velocity changes seem to be slightly underes-
timated by DWI due to the decorrelation.

The decorrelation coefficient is a clear indicator that the shape

of the wave has changed, in which case, the value of the time
stretching (and therefore the wave velocity change) is not reliable.
Fig. 9. Snapshot of a signal and definition of the time windows used for the CWI When wave velocity changes in the media are uniform, time
and the DWI methods in the case of the bending test. stretching gives a reliable estimate of the wave velocity change.
48 A. Deraemaeker, C. Dumoulin / Construction and Building Materials 194 (2019) 42–50

For (i) and (ii), the results have shown the excellent sensitivity of
the damage indicator to the onset of cracking, the indicator is able
to detect damage before cracks are visible on the outer surface. For
(iii), we have found that because of the higher levels of stresses
involved in compression tests, the acoustoelastic effect [10] was
affecting the damage index. An alternative indicator was therefore
proposed which consists in monitoring the change of amplitude of
the first peak only. In much the same way, we propose here to
apply the damage index on the stretched rather than the initial sig-
nal using the time stretching technique described in Sec. 4.2. This
allows to get rid of the sensitivity to the change of the wave veloc-
ity and focus on the change of amplitude and shape of the first
peak. The damage index DI is now expressed as
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uR t2  
u t Sj ðtð1
max ÞÞ
S0 ðtÞ 2 dt
DIj ¼ t 1
R t2 2 ð7Þ
t1 0
S ðt Þdt

where Sj ðt Þ corresponds to the amplitude of the damaged signal,

S0 ðtÞ is the amplitude of the healthy signal, t 1 is the time of arrival
Fig. 11. Evolution of the decorrelation as function of the time (bending test). of S0 , and t 2
t 1 corresponds to the duration of the first half-period.
This is represented on Fig. 12 where an example of a healthy and a
signal with cracking are presented. max corresponds to the value of
In our tests, we see that when using CWI, the wave shape is time stretching which maximizes the correlation coefficient CC
altered well before damage occurs. The CODA contains wave defined in (6).
packets which have travelled in the whole concrete specimen. Fig. 13 shows that if no time stretching is applied to the mea-
As the stress state is not uniform due to the nature of the test sured signals, the DI is affected by these small velocity changes
(bending), the acoustoelastic effect will have a non-uniform (see the small increase of the DI around point 1 which corresponds
impact on the CODA, resulting in a decorrelation. Other factors to the time when the velocity change is the highest according to
such as stress concentration in the areas close to the boundary the DWI time stretching presented in Fig. 10). Applying time
conditions have a similar effect. When using DWI, the wave that stretching allows to get rid of the small velocity changes due to
is recorded and used to compute the decorrelation has only trav- operational perturbations, at the cost of a slightly lower sensitivity
elled in a restricted area around the transducers, and we notice to the damage.
that the correlation is very good before the damage occurs. As In Fig. 14 we compare the DI corrected with time stretching to
the damage induces a local change of wave velocity between the evolution of the amplitude of the first peak of the wave previ-
the transducers, the decorrelation starts to increase with its ously proposed in [30] as a damage indicator robust to environ-
appearance. mental factors such as the acoustoelastic effect. As the figure
One can conclude that before the damage occurs, DWI can be
used with confidence to track accurately very small (up to 0.3%
in our test) velocity changes. We believe that these velocity
changes are mainly due to the acoustoelastic effects. The focus of
this paper is however not to understand deeply where these
changes are coming from, but to propose a method to filter out
their effects.

4.3. Damage monitoring under changing operational conditions

For real-time monitoring, it is important to be able to distin-

guish between the effects of damage and operational conditions.
As the wave velocity varies due to the appearance of damage as
well as changing operational conditions, it is not a robust indicator
of damage. The previously used damage indicator, the early wave
damage index, has also been shown to be sensitive to operational
conditions [30]. In this paper, we propose to apply time stretching
with DWI in order to filter out operational conditions before com-
puting the damage index.
The time-of-flight and the first wave amplitude are known to be
sensitive to cracking. We have therefore proposed in [29] to define
a damage indicator which focuses on the early wave in the signal.
This damage indicator has been used to monitor cracking in differ-
ent laboratory setups consisting in:

(i) three points bending tests on concrete beams [29,31],

(ii) pull-out tests on an anchorage in a large concrete block
[29,33], Fig. 12. Example of a healthy and a damaged signal and definition of the quantities
(iii) compression tests on a cylinder [30]. used for the computation of the damage index (with and without time stretching).
 A. Deraemaeker, C. Dumoulin / Construction and Building Materials 194 (2019) 42–50 49

cracking in the middle of the beam. Three different techniques

were analyzed in order to track small relative velocity changes.
The time of flight has a low resolution (around 1%) which can be
improved to 0.1% with denoising if the number of measurements
is sufficient. The CWI approach has been shown to be too sensitive
to the non-uniform stress field and effects of boundary conditions
in the test. While its accuracy is very high (around 2:10
5 ), it is not
applicable for in situ monitoring of the wave velocity changes on
real structures. As it is focused only on the early wave arrive, the
DWI has a slightly lower accuracy (around 2:10
4 ) but is much
more robust with respect to the loading conditions and changing
boundary conditions. We have shown that it is able to track accu-
rately small velocity changes before the damage appears.
DWI can therefore be used to stretch the recorded signals and
correct for small wave velocity changes due to operational condi-
tions. Based on this finding, we have defined a damage indicator
which is made robust to operational changes thanks to time
stretching using DWI. The velocity changes in our tests are thought
Fig. 13. Evolution of the damage indicator as function of the time (DWD: direct
to be due to the effect of applied stresses (only acoustoelastic
wave decorrelation, CWD: coda wave decorrelation). effect), and not to environmental factors such as temperature or
humidity. The applicability of the technique in changing environ-
mental conditions remains to be tested, but as environmental
changes are likely to produce more or less uniform changes of
material properties, the velocity changes are also likely to be uni-
form so that time stretching is applicable to compensate for these
effects and differentiate them from cracking.
In our test, the corrected damage indicator is able to track effi-
ciently the appearance of damage while being insensitive to chang-
ing operational conditions. The results obtained have been
compared to the previously proposed damage indicator based on
the first peak amplitude, which shows very similar results. Both
approaches are therefore seen as very efficient for automated long
term monitoring of cracking in concrete structures using embed-
ded piezoelectric transducers.
One important issue which needs further investigation is the a
priori determination of the final time t f to be used in DWI in order
not to be sensitive to changing boundary conditions and non-
uniform stress fields. This will be investigated in more details in
the future.

Fig. 14. Evolution of the damage indicator, and the first peak amplitude as function Conflict of interest
of the time (bending test).

There is no conflict of interest.

shows, the two approaches lead to very similar results. The choice
between one or the other should be mainly dictated by the compu- Acknowledgments
tational efforts involved for the computation of each indicator.
This study was supported by the Belgian National Fund for Sci-
entific Research (FRS-FNRS) and by the Fonds David et Alice Van
5. Conclusions Buuren.

This paper presented a methodology for lifelong monitoring of References

concrete structures. The technique is based on the use of pairs of
low-cost piezoelectric transducers which are embedded inside [1] A. Saul, Principles underlying the steam curing of concrete at atmospheric
pressure, Magazine Concr. Res. 6 (2) (1951) 127–140.
the concrete structure during the construction phase. A high- [2] T. Yikici, H. Chen, Use of maturity method to estimate compressive strength of
frequency pulse is sent to the emitter and the receiver records mass concrete, Constr. Build. Mater. 95 (2015) 802–812.
the wave after propagation. Based on the recorded signals at regu- [3] EN 12504-4:2004 (E). Testing concrete. Determination of ultrasonic pulse
velocity. European Standard 2004;.
lar time intervals, information about the state of the microstruc- [4] ASTM C597, Standard Test Method for Pulse Velocity Through Concrete.
ture of concrete can be obtained. The efficiency of the method American Society for Testing and Materials, West Conshohocken, PA, USA
has been demonstrated through laboratory tests on two small con- 2016;:1–4. doi:10.1520/C0597-09.
[5] F. Saint-Pierre, A. Philibert, B. Giroux, P. Rivard, Concrete quality designation
crete beam specimen.
based on ultrasonic pulse velocity, Constr. Build. Mater. 125 (2016) 1022–
At early age, we have shown that it was possible to track the P- 1027.
wave velocity change from the liquid to the solid state, and to use [6] BS 1881 – part 203. Testing concrete. Recommendations for measurement of
it to estimate the evolution of mechanical properties in real-time. velocity of ultrasonic pulses in concrete. British Standard 1986.
[7] J. Carette, S. Staquet, Monitoring the setting process of eco-binders by
For long-term monitoring, we have tested one of the sample beams ultrasonic p-wave and s-wave transmission velocity measurement: mortar
under a three-point bending test in order to induce progressive vs concrete, Constr. Build. Mater. 94 (2016) 32–41.
50 A. Deraemaeker, C. Dumoulin / Construction and Building Materials 194 (2019) 42–50

[8] J. Bogas, M. Gomes, A. Gomes, Compressive strength evaluation of structural Health Monit. 16 (2) (2017) 202–214, https://doi.org/10.1177/
lightweight concrete by non-destructive ultrasonic pulse velocity method, 1475921716670573.
Ultrasonics 53 (2013) 962–972. [28] C. Dumoulin, G. Karaiskos, J. Carette, S. Staquet, A. Deraemaeker, Monitoring of
[9] E. Larose, S. Hall, Monitoring stress related velocity variation in concrete with a the ultrasonic P-wave velocity in early-age concrete with embedded
2  10(
5) relative resolution using diffuse ultrasound, J. Acoust. Soc. Am. 125 piezoelectric transducers, Smart Mater. Struct. 21 (4) (2012) , https://doi.org/
(4) (2009) 1853–1856, https://doi.org/10.1121/1.3079771. 10.1088/0964-1726/21/4/047001 047001.
[10] I. Lillamand, J.F. Chaix, M.A. Ploix, V. Garnier, Acoustoelastic effect in concrete [29] C. Dumoulin, G. Karaiskos, J.Y. Sener, A. Deraemaeker, Online monitoring of
material under uni-axial compressive loading, NDT & E Int. 43 (8) (2010) 655– cracking in concrete structures using embedded piezoelectric transducers,
660, https://doi.org/10.1016/j.ndteint.2010.07.001. Smart Mater. Struct. 23 (11) (2014) , https://doi.org/10.1088/0964-1726/23/
[11] E. Larose, A. Obermann, A. Digulescu, T. Planès, J.F. Chaix, F. Mazerolle, et al., 11/115016 115016.
Locating and characterizing a crack in concrete with diffuse ultrasound: a four- [30] C. Dumoulin, A. Deraemaeker, Real-time fast ultrasonic monitoring of concrete
point bending test, J. Acoust. Soc. Am. 138 (1) (2015) 232–241, https://doi.org/ cracking using embedded piezoelectric transducers, Smart Mater. Struct. 26
10.1121/1.4922330. (10) (2017) , https://doi.org/10.1088/1361-665X/aa765e 104006.
[12] Y. Zhang, T. Planès, E. Larose, A. Obermann, C. Rospars, G. Moreau, Diffuse [31] C. Dumoulin, G. Karaiskos, A. Deraemaeker, 8 – Monitoring of crack
ultrasound monitoring of stress and damage development on a 15-ton propagation in reinforced concrete beams using embedded piezoelectric
concrete beam, J. Acoust. Soc. Am. 139 (4) (2016) 1691–1701, https://doi. transducers, in: M. Ohtsu (Ed.), Acoustic Emission and Related Non-
org/10.1121/1.4945097. Destructive Evaluation Techniques in the Fracture Mechanics of Concrete.
[13] P. Frojd, P. Ulriksen, Continuous wave measurements in a network of Woodhead Publishing Series in Civil and Structural Engineering, Woodhead
transducers for structural health monitoring of a large concrete floor slab, Publishing, Oxford, 2015, pp. 161–175. ISBN 978-1-78242-327-0.
Struct. Health Monit. 15 (4) (2016) 403–412, https://doi.org/10.1177/ [32] C. Dumoulin, A. Deraemaeker, Design optimization of embedded ultrasonic
1475921716642139. transducers for concrete structures assessment, Ultrasonics 79 (2017) 18–33,
[14] L. Qin, Z. Li, Monitoring of cement hydration using embedded piezoelectric https://doi.org/10.1016/j.ultras.2017.04.002.
transducers, Smart Mater. Struct. 17 (2008) 055005. [33] C. Dumoulin, G. Karaiskos, A. Deraemaeker, Concrete monitoring using
[15] Q. Kong, S. Hou, Q. Ji, Y. Mo, G. Song, Very early age concrete hydration embedded piezoelectric transducers, 1 ed., in: M. Beer, I. Kougioumtzoglou,
characterization monitoring using piezoceramic based smart aggregates, E. Patelli, I.K. Au (Eds.), Encyclopedia of Earthquake Engineering, Springer,
Smart Mater. Struct. 22 (2013) 085025. Berlin Heidelberg, Berlin, Heidelberg, 2015, pp. 1–12, https://doi.org/10.1007/
[16] H. Gu, G. Song, H. Dhonde, Y. Mo, S. Yan, Concrete early-age strength 978-3-642-36197-594-1. ISBN 978-3-642-35345-1.
monitoring using embedded piezoelectric transducers, Smart Mater. Struct. 15 [34] M. Ohtsu, editor. Acoustic Emission and Related Non-Destructive Evaluation
(2006) 1837–1845. Techniques in the Fracture Mechanics of Concrete: Fundamentals and
[17] W. Dansheng, Z. Hongping, Monitoring of strength gain of concrete using Applications. Woodhead Publishing Series in Civil and Structural
embedded PZT impedance transducer, Constr. Build. Mater. 25 (2011) 3703– Engineering: Number 57; Cambridge: Elsevier; 2015. ISBN 9781782423270.
3708. DOI: 10.1016/B978-1-78242-327-0.01001-1.
[18] G. Song, H. Gu, Y. Mo, T. Hsu, H. Dhonde, Concrete structural health monitoring [35] C. Grosse, M. Ohtsu, Acoustic Emission Testing, Springer, Berlin Heidelberg,
using embedded piezoceramic transducers, Smart Mater. Struct. 16 (2007) Berlin, Heidelberg, 2008, ISBN 978-3-540-69895-1, https://doi.org/10.1007/
959–968. 978-3-540-69972-9.
[19] X. Zhao, H. Li, D. Du, J. Wang, Concrete structure monitoring based on [36] J.H. Kurz, C.U. Grosse, H.W. Reinhardt, Strategies for reliable automatic onset
built-in piezoelectric ceramic transducers, Proc. SPIE 6932 (2008) time picking of acoustic emissions and of ultrasound signals in concrete,
693208. Ultrasonics 43 (7) (2005) 538–546, https://doi.org/10.1016/j.ultras.2004.
[20] V. Annamdas, Y. Yang, C. Soh, Impedance based concrete monitoring using 12.005.
embedded PZT sensors, Int. J. Civil Struct. Eng. 1 (3) (2010) 414–424. [37] I. Daubechies, Ten lectures on wavelets, Soc. Ind. Appl. Math. (1992), https://
[21] W. Liao, J. Wang, G. Song, H. Gu, C. Olmi, Y. Mo, et al., Structural health doi.org/10.1137/1.9781611970104. ISBN 978-0-89871-274-2.
monitoring of concrete columns subjected to seismic excitations using [38] B. Delsaute, C. Boulay, J. Granja, J. Carette, M. Azenha, C. Dumoulin, et al.,
piezoceramic-based sensors, Smart Mater. Struct. 20 (2011) 125015. Testing concrete E-modulus at very early ages through several techniques: an
[22] Q. Feng, Q. Kong, L. Huo, G. Song, Crack detection and leakage monitoring on inter-laboratory comparison, Strain 52 (2) (2016) 91–109, https://doi.org/
reinforced concrete pipe, Smart Mater Struct 24 (2015) 115020. 10.1111/str.12172.
[23] D. Zou, T. Liu, Y. Huang, F. Zhang, C. Du, B. Li, Feasibility of water seepage [39] C. Dumoulin, A. Deraemaeker, From early age assessment of concrete
monitoring in concrete with embedded smart aggregates by P-wave travel properties to crack detection using embedded ultrasonic transducers, 2nd
time measurement, Smart Mater Struct 23 (2014) 067003. International RILEM/COST Conference on Early Age Cracking and Serviceability
[24] Q. Kong, Q. Feng, G. Song, Water presence detection in a concrete crack using in Cement-based Materials and Structures – EAC2, vol. 1, RILEM, Brussels
smart aggregates, Int. J. Smart Nano Mat. 6 (3) (2015) 149–161. (Belgium), 2017, pp. 137–142.
[25] T. Liu, Y. Huang, D. Zou, J. Teng, B. Li, Exploratory study on water seepage [40] R. Snieder, J. Scales, Time-reversed imaging as a diagnostic of wave and
monitoring of concrete structures using piezoceramic based smart aggregates, particle chaos, Phys. Rev. E 58 (5) (1998) 5668–5675, https://doi.org/10.1103/
Smart Mater. Struct. 22 (6) (2013), https://doi.org/10.1088/0964-1726/22/6/ PhysRevE.58.5668.
065002. 065002. [41] A. Tourin, M. Fink, A. Derode, Multiple scattering of sound, Waves Random
[26] Z. Li, L. Qin, S. Huang, Embedded piezo-transducer in concrete for property Media 10 (4) (2000) R31–R60, https://doi.org/10.1088/0959-7174/10/4/201.
diagnosis, J. Mater. Civil Eng. 21 (11) (2009) 085025. [42] D.P. Schurr, J.Y. Kim, K.G. Sabra, L.J. Jacobs, Damage detection in concrete using
[27] T. Liu, D. Zou, C. Du, Y. Wang, Influence of axial loads on the health monitoring coda wave interferometry, NDT & E Int. 44 (8) (2011) 728–735, https://doi.org/
of concrete structures using embedded piezoelectric transducers, Struct. 10.1016/j.ndteint.2011.07.009.