ARTICLE IN PRESS

NDT&E International 41 (2008) 258–264

www.elsevier.com/locate/ndteint

Interferometric radar vs. accelerometer for dynamic monitoring of large

structures: An experimental comparison
Massimiliano Pieraccinia,, Matteo Fratinia, Filippo Parrinia, Carlo Atzenia, Gianni Bartolib
a
Department of Electronics and Telecommunications, University of Florence, via Santa Marta 3, 50139 Florence, Italy
b
Department of Civil Engineering, University of Florence, via Santa Marta 3, 50139 Florence, Italy
Received 11 July 2007; received in revised form 22 November 2007; accepted 23 November 2007
Available online 28 November 2007

Abstract

Interferometric radar has been recently proposed as a measurement instrument for dynamic testing/monitoring of large structures,
such as bridges, towers, buildings, and dams, which is currently performed by networks of accelerometers. In this paper, the authors
report a direct comparison between the two measurement techniques (radar interferometer vs. accelerometers) both employed during a
field test on a bridge. As different quantities, displacement and acceleration are measured by the two techniques, a preliminary discussion
about signals and noise has been necessary. Finally, the experimental results are critically discussed.
r 2007 Elsevier Ltd. All rights reserved.

Keywords: Radar; Remote sensing; Interferometry; Dynamic monitoring; Accelerometer; Structural test

1. Introduction sensitive to dust and abrupt changes of environmental

conditions [4].
Dynamic testing of large architectural and mechanical Recently, the authors have developed a remote micro-
structures is currently implemented by networks of wave sensor able to provide displacement measurements
accelerometers [1–3]. Such sensors are accurate and with sub-millimeter accuracy and a sampling rate high
reliable, but need to be positioned in close contact with enough to track the transient movements of an architec-
the monitored structure. The installation of sensors can tural structure. In effect, a typical architectonic structure
necessitate maintaining the structure out of service for a (like a bridge, a tower, a building, etc.) has natural
considerable amount of time. Furthermore, monitoring of frequencies in the order of a few hertz and displacements
large structures can give rise to accessibility problems, up to several centimeters.
often requiring the use of costly and cumbersome The sensor, an interferometric continuous-wave step-
scaffolding. In a number of situations, placing of contact frequency (CWSF) radar, has been tested on bridges [5–7]
sensors may not be possible at all: this is the case, for and towers [8,9].
example, in buildings with symptoms of impending As the accelerometers are the most popular instruments
collapse, after a seismic shock or a blast. for dynamic testing of large structure, in this paper, a direct
Laser sensors are a possible non-contact technique for comparison between the interferometric radar and con-
monitoring large structures, but a laser sensor can detect ventional accelerometers is reported. With this aim, a field
the displacement of a single selected point of the structure, test on a bridge has been carried out, and the experimental
while the whole deformation pattern is often required, results are critically discussed.
furthermore laser sensors have been proven often unprac-
tical for in-field applications, because they are very
2. Interferometric radar and accelerometer network

Corresponding author. Tel.: +39 055 479 6273. The interferometric sensor is a CWSF coherent radar,
E-mail address: massimiliano.pieraccini@unifi.it (M. Pieraccini). able to provide images of the illuminated scenario up to a

0963-8695/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ndteint.2007.11.002
 ARTICLE IN PRESS
M. Pieraccini et al. / NDT&E International 41 (2008) 258–264 259

R R
h
z
bridge
radar

Fig. 1. Radar measurement geometry (h—bridge height from radar antennas, R—distance of the bridge section from the radar, DR—variation of the
distance of the bridge section from the radar, Dz—variation of the bridge section vertical position).

2 km range with 0.5 m resolution, and with an image rate

up to 100 Hz. The instrument has been designed with an
electronic stability able to provide a better than 100 mm
accuracy. The technical details of this instrument are
reported in literature [6,7] and are not repeated here. This
instrument is able to detect the displacements by measuring
the rotation, in the complex plane, of the phasor related to
the backscattered signal of a single resolution cell.
The detected phasor is the summation of all the phasors
related to each scatterer in that resolution cell. It should
be noted that the radar has only range resolution or, in
other words, it is able to distinguish object on distance
basis only. The resolution cell length is equal to the range
resolution of the instrument, while its angular width is
constrained only by the antenna’s lobe. Only if the
scatterers of a single resolution cell carry out a rigid shift,
the rotation angle of the summation phasor is linearly
related to the displacement along the radar line of view. Fig. 2. The radar equipment.
Rotation or deformation of the set of scatterers can result
in a measurement error that has to be evaluated in
operating conditions. Inc. They are reported to have a 70.5 dB frequency
As this radar is a mono-dimensional system, able to response from 1 Hz to 4 kHz. The resonant frequency of
distinguish objects on the basis of their measured ranges, the sensors is variable in function of the mounting
but not of angles, in order to acquire an entire structure, it configuration. In our case it is reported to be much greater
is necessary to install it in such a way that different zones of than 22 kHz.
the structure appear at different ranges. This requirement is The signal from each accelerometer was amplified by a
only obtainable on structures with longitudinal extension flat-response amplifier with an amplitude gain of 100. The
such as bridges, towers and similar, putting the radar view signals have been logged with a notebook and a 16 bit
cone in a skew position with respect to the structure itself. acquisition device part of 9800 Series produced by Data
It should be also noted that the radar is able to detect the Translation Inc.
component DR of the displacements along its line of sight
only. Therefore, it is necessary to multiply the measured 3. Signal analysis
displacement by a geometrical factor to obtain the effective
displacement. By assuming Dz as a vertical displacement, In order to proceed to a direct comparison between
this factor is the ratio between the distance from the radar radar and accelerometers, a preliminary question has to be
R of the point where the displacement is detected and the discussed: radar measures displacements, while acceler-
height h of the same point with respect to the radar antenna ometers measure accelerations. These two quantities are
horizontal plane (see Fig. 1). not homogenous, but linked by integral/differential rela-
Fig. 2 shows the radar equipment at work during the tionships. It is thus necessary to integrate the accelera-
test. tion data twice in order to make a comparison of
The accelerometer network employed in this experiment displacement, or to differentiate the radar displacement
was conventional. It was composed by three sensors data twice to make a comparison of acceleration.
connected to a central data acquisition system through Furthermore, it is possible to integrate the accelerometer
three signal amplifiers and coaxial cables. The model of the data once and differentiate the radar ones once to compare
employed accelerometers is the 353B34  PCB Piezotronics data in speed.
 ARTICLE IN PRESS
260 M. Pieraccini et al. / NDT&E International 41 (2008) 258–264

In particular, the following points have to be taken into z DT

K i ðzÞ ¼ DT ) H i ðoÞ ¼ , (3)
account: z1 2 sinðoDT=2Þ
(1) The typical signal spectrum of a solicited large
  2
structure consists of a number of peaks, related to z 2 DT
vibrational modes. These peaks are not uniformly dis- K ii ðzÞ ¼ DT ) H ii ðoÞ ¼ . (4)
z1 2 sinðoDT=2Þ
tributed in frequency, but typically the peaks at lowest
frequency (the first modes) contain more than 95% of the The ratio between signal spectrum and noise spectrum is
spectral power. On the other hand, the radar noise can be not altered by differentiation or integration, but the overall
considered white and Gaussian while the accelerometers signal power to noise power ratio could be, as the signal
one depends on the employed sensors. and noise do not have the same spectral distribution and
(2) The accelerometers have a band-pass frequency the filter spectral response is not flat in the band of interest.
response. Furthermore, data from real sensors can be In particular, taking into account what is stated above in
affected by a measurement offset, caused by both the point (1), it is possible to calculate the SNR of operations
sensor itself and the data acquisition device analog front from (1) to (4) in case of known noise spectral distribution
end. It is therefore necessary to filter both radar and N(o) and single sinusoid with a fixed amplitude A, as a
accelerometer data with the same band-pass filter. function of its pulsation oS:
(3) The radar and the accelerometers sample the structure 1 2 2
2A HðoS Þ
under test with a different sampling rate. The accelerometer SNRðoS Þ ¼ R oh 2 2
; ol ooS ooh . (5)
ol NðoÞ HðoÞ do
sampling frequency is significantly higher than the double of
the maximum expected oscillation frequency of the structure. The values oh and ol represent the upper and lower cut-
This high sampling frequency is not only used in order to off pulsation of the band-pass filter. With the aim of
avoid aliasing but in order to lower the quantization noise verifying the formulas (1)–(4) and the hypothesis of white
too. This sampling frequency is usually greater than that noise for the radar prototype, a specific measurement has
obtainable using the CWSF radar technique. Therefore, it is been performed in Laboratory A wall has been used as
necessary to down-sample the accelerometer data to equalize a stable target for the radar in order to sample the noise.
its sampling frequency to the radar one. It should be noted Fig. 3 shows the theoretical and measured acceleration
that this down-sampling do not degrade the signal-to-noise noise spectrum after a double differentiation of the
ratio (SNR) of the accelerometer data, because they have displacement. The noise plot is obtained by calculating
been previously filtered using the same band-pass filter that the average of the amplitude of 20 different spectra, after
has been used on radar data. double differentiation. As visible, it is a filter shaped white
(4) Differentiation of the radar data permits comparison noise as predicted by theory.
with acceleration data, but integration of the accelerations to In order to study how the SNR changes in case of
obtain speed and displacement could be a problem as two sinusoidal signal in function of its pulsation oS, it is
integration constants are needed, but not known. Filtering necessary to know the noise spectrum distribution N(o). In
out low frequencies solves this problem and enables an easy this situation, the following noise figures can be defined
comparison of all the three physical variables (displacement, and calculated as
speed and acceleration). The integration constants are no R oh 2 2
longer needed and they have to be assigned a value of zero. SNRIN 1 ol NðoÞ H x ðoÞ do
F x ðoS Þ ¼ ¼ R o ,
(5) Derivation and integration operate like spectral filters SNROUT H x ðoS Þ2 h 2
ol NðoÞ do
that amplify or attenuate different zones of the spectrum of
ol ooS ooh , ð6Þ
the signal and of the noise. A filter is usually mathemati-
cally described using its transfer function which is the Z where x can be d, dd, i, ii (respectively, single and double
transform of its pulse response in time, in case of a sampled differentiation, and single and double integration).
system. In the following equations, Kd(z) is the Z transform Fig. 4 shows the plots of the noise figures Fii(o) (Curve
of the differentiation (1), Kdd(z) of the double differentia- A), Fdd(o) (Curve C) and |Fd(o)/Fi(o)| (Curve B),
tion (2), Ki(z) of the integration (3), and Kii(z) of the double expressed in dB scale, as a function of o, in the case of
integration (4); DT is the sampling period and H(o) the oh ¼ 30 Hz, ol ¼ 1 Hz and DT ¼ 1/75 s. It is important to
angular frequency response: note that, for calculating the noise figures, it is important to
  know only the shape of N(o) and not its absolute value. As
1 z1 2 oDT the accelerometer data have to be integrated to obtain the
K d ðzÞ ¼ ) H d ðoÞ ¼ sin , (1)
DT z DT 2 other two physical variables, in order to plot the Curve A,
the noise spectrum N(o) has been retrieved from accel-
 
1 z1 2 erometer datasheet. As the radar data have to be
K dd ðzÞ ¼ ) H dd ðoÞ differentiated to obtain the other two physical variables,
DT z
  2 in order to plot the Curve C, the noise spectrum N(o) has
2 oDT been considered flat. Similar considerations can be done
¼ sin , ð2Þ
DT 2 regarding Curve B. Observing Fig. 4 it is possible to see
 ARTICLE IN PRESS
M. Pieraccini et al. / NDT&E International 41 (2008) 258–264 261

140

120

Acceleration [mm/sec2] 100

80

60

40

0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Frequency [Hz]

Fig. 3. Measured and theoretical (full line) double differentiated noise spectrum of the displacement of a fixed scatterer acquired by radar.

45 Curve A
40
35
30
Noise Figures [dB]

25 Curve B
20
15
10
5
0
Curve C
-5
-10

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
B

Frequency [Hz]
ne
Zo

Zone A Zone C Zone D Zone E

Fig. 4. Noise figures and zones (Curves A, B, and C represent the noise figures, Zones A and B are that in which a displacement comparison is convenient,
Zone C is that in which a speed comparison is convenient, Zones D and E are that in which an acceleration comparison is convenient).

that, depending on frequency, the derivative or integral can C, and D) both the double integration and the double
lower or enhance the SNR. With the aim of comparing differentiation lower the SNR. However, the plot of |Fd(o)/
data with the best SNR, it is necessary to choose the correct Fi(o)| (Curve B) helps to choose the better physical
physical variable (displacement, speed, acceleration). variable to be used. Where this curve remains under the
A double integration enables displacement comparison other two (o3popo4 (Zone C)), it is better to compare
and improves the SNR of accelerometer data for pulsation using speed; for o1popo3 (Zone B) it is better to
ooo1 (Zone A, Fii(o1) ¼ 0 dB). Therefore, the compar- compare displacement, and finally for o2popo4 (Zone
ison of displacement is the best choice in this pulsation D) it is better to compare acceleration.
band, as the radar data are directly compared with
accelerometer data cleaned up by a double integration. A 4. The in-field test
double differentiation improves the SNR of radar data for
pulsation o4o2 (Zone E, Fdd(o2) ¼ 0 dB). Therefore, the The experimental campaign was conducted on a cable-
comparison of acceleration is the best choice in this stayed steel pedestrian bridge, located in the village of
pulsation band, as the accelerometer data are directly Poggibonsi (Siena, Italy), with a central span of about
compared with radar data cleaned up by a double 45 m. Eight pairs of stays symmetrically support the steel
differentiation. In the inner band o1popo2 (Zones B, deck, which presents a width of about 2.50 m, and are
 ARTICLE IN PRESS
262 M. Pieraccini et al. / NDT&E International 41 (2008) 258–264

restrained to two inclined pylons. The relatively small Fig. 6, the position of the three accelerometers, labeled with
stiffness of the structure allowed a clear identification of #1, #2, and #3, are also reported. The third accelerometer
the bridge motion due to external dynamical actions. has been only used to check the consistency of the
Generally speaking, a bridge can move showing flexional acquisition and it has not been used in the comparison.
(vertical deflection along longitudinal axis) and torsional As the maximum oscillation frequency of the structure
(angular deflection on longitudinal axis) oscillation modes. was expected to be within 20 Hz, the accelerometers were
However, the presented work is aimed to compare sampled with a sampling frequency of 1 kHz with an anti-
the accelerometer data with radar one in case of flexional aliasing analog low-pass filter with a cut frequency of
mode only. A radar data processing, able to extract 400 Hz.
correct information on torsional ones, is currently under To proceed to the comparison, a solicitation able to
development. reduce the amplitude of the torsional components has been
In this experiment, the radar was positioned at the basis applied to the bridge. Indeed the structure has been
of one of the two pillars as shown in Fig. 6 in which a dynamically excited by a man walking and running over
drawing of the bridge and the radar position, labeled with the central longitudinal axis of it.
#R, are reported. The radar was configured to obtain a In order to synchronize the radar and the accelero-
range resolution of 40 cm and a sampling frequency of meter acquisitions, the radar has been provided with a
75 Hz. trigger-out signal that was acquired by the accelerometer
In order to acquire separately the flexional and torsional data logger.
oscillation components of a bridge section, two acceler- As the radar sampling frequency was configured at 75 Hz
ometers are usually installed on the extremities of it. the Nyquist frequency was at 37.5 Hz. However, the data
Torsional oscillations can be obtained as a differential has been considered valid in the lower 80% of the Nyquist
mode component of the two signals and flexional as the band only or, in other words, between 0 and 30 Hz. On the
common mode one. In other words, if flexional modes only other hand, the employed accelerometers have a frequency
are excited the two accelerometers report the same signals, response starting at 1 Hz. According to the discussion
if torsional modes only are excited the two accelerometers reported in the third paragraph, the two sets of data have
report opposite signals. been band-passed in the 1–30 Hz band.
In this experiment three accelerometers were placed on Following the theory discussed in third paragraph, the
the structure, in positions easily detectable by the radar. As three zones in which is convenient to perform a comparison
the radar is able to distinguish the displacement of points at of displacement, speed, and acceleration are, respectively,
different distances, the bin in the radar image related to the 1–3.5 Hz (Zones A and B), 3.5–10.5 Hz (Zone C), and
accelerometer position has been identified by taking into 10.5–30 Hz (Zones D and E). The frequency of the main
account the geometry of the bridge and the position of the peak of the acquired signal was about 2.1 Hz. The
radar. maximum frequency component was instead 2.9 Hz.
Two accelerometers were placed on the extremities of a A complete spectrum of an accelerometer is reported in
same transversal section of the bridge. The third was Fig. 5. Comparing this figure with Fig. 4 it is possible to
instead placed on a different section, in a symmetrical observe that the all the signal is contained in the Zones A
position with respect to the center of the bridge long- and B. Therefore, in this experiment, the best comparison
itudinal extension. Both sections are 15 m away from the between radar and accelerometer could be carried out
pylons, moving toward the symmetry axis of the bridge. In using displacement variable.

0.08

0.07

0.06
Displacement [mm]

0.05

0.04

0.03

0.02

0.01

0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Frequency [Hz]

Fig. 5. Spectrum of oscillation of the selected bridge section as acquired by radar.

 ARTICLE IN PRESS
M. Pieraccini et al. / NDT&E International 41 (2008) 258–264 263

#2 #3

#1

View Cone

#R #2 #1 #3

Fig. 6. Drawing of the bridge, the radar (#R), the view cone and the accelerometers (#1, #2, and #3).

1.5

1
Displacement [mm]

0.5

-0.5

-1

-1.5
35 36 37 38 39 40 41 42 43 44 45
Time [sec.]

Fig. 7. Displacement comparison within 10 s between radar data (full line) and accelerometers (dotted line).

Fig. 7 shows the radar signal and one of the two 1.12 mm which is considerably greater than 0.122 mm,
accelerometers, after band-pass filtering and double inte- confirming that the agreement between radar and accel-
gration during 10 s of total acquisition time. erometer is better in displacement than in acceleration, as
The agreement is very good. The standard deviation of the theoretically justified.
differences between radar and accelerometers, calculated over
the entire acquisition of 160 s, results to be 0.122 mm. 5. Conclusion
Above in this section we demonstrated that the best way
to compare the two instrumentations is in displacement, This paper has reported a direct comparison between
nevertheless the accelerometers are the most popular displacement measurements performed by a radar inter-
sensors for dynamic structural tests and technicians are ferometer and a network of accelerometers, both employed
more familiar with these kinds of data, even if often they during a field test on a bridge. As the two instruments
integrate them to obtain the displacements. Therefore, for measure different quantities, displacement, and accelera-
sake of completeness, Fig. 8 shows the comparison of tion, the choice of the signals to be compared and their
acceleration between of the two instruments. The standard processing have been preliminarily discussed.
deviation of the differences between radar and acceler- In effect, depending on the main spectral component of
ometers, calculated over the entire acquisition of 160 s, the signal and on the spectral shape of the noise, the best
results to be 194.7 mm/s2. agreement between radar and accelerometers can be in
In order to compare these two standard deviations in displacement, speed, or acceleration. For the specific case,
different units, it can be interesting to convert the the displacement has been found the best variable.
acceleration to displacement by supposing, very roughly, Taking into account these considerations, we can
that all the signal energy is contained in its strongest conclude that, almost in the experimental conditions
component at 2.1 Hz. Under this hypothesis, the accelera- described in this paper, the standard deviation between
tions can be converted to displacement by dividing it by the radar and accelerometers is of 0.122 mm. In other words,
square of the signal pulsation. The obtained value is the two instruments give results in agreement with an error
 ARTICLE IN PRESS
264 M. Pieraccini et al. / NDT&E International 41 (2008) 258–264

500

400

300

200
Accel [mm/sec2]

100

-100

-200

-300

-400

-500
35 35.5 36 36.5 37 37.5 38 38.5 39 39.5 40
Time [sec.]

Fig. 8. Acceleration comparison within 5 s between radar data (full line) and accelerometers (dotted line).

bar in the order of one-tenth of millimeter, but the [3] Lee JW, Kim JD, Yun CB, Yi JH, Shim JM. Health-monitoring
interferometer has the unique advantage to operate with- method for bridges under ordinary traffic loadings. J Sound Vib 2002;
out contact by acquiring simultaneously the whole 257:247–64.
[4] Nassif HH, Gindy M, Davisa J. Comparison of laser Doppler
deformation of the deck. vibrometer with contact sensors for monitoring bridge deflection and
vibration. NDT&E Int 2005;38:213–8.
Acknowledgment [5] Pieraccini M, Parrini F, Fratini M, Atzeni C, Spinelli P, Micheloni M.
Static and dynamic testing of bridges through microwave interfero-
metry. NDT&E Int 2007;40:208–14.
The equipment used in this work has been designed and [6] Pieraccini M, Fratini M, Parrini F, Macaluso G, Atzeni C. High-speed
constructed with the support of IDS—Ingegneria dei CW step-frequency coherent radar for dynamic monitoring of civil
Sistemi SpA, Pisa (Italy), and it remains the property of engineering structures. Electron Lett 2004;40:907–8.
this company. [7] Pieraccini M, Fratini M, Parrini F, Atzeni C. Dynamic monitoring of
bridges using a high speed coherent radar. IEEE Trans Geosci Remote
Sensing 2006;44:3284–8.
References [8] Pieraccini M, Fratini M, Parrini F, Pinelli G, Atzeni C. Dynamic
survey of architectural heritage by high-speed microwave interfero-
[1] Hearn G, Testa RB. Modal analysis for damage detection in metry. IEEE Geosci Remote Sensing Lett 2005;1:28–30.
structures. J Struct Eng 1991;117:3042–63. [9] Pieraccini M, Parrini F, Dei D, Fratini M, Atzeni C. Dynamic
[2] Shieh J, Huber JE, Fleck NA, Ashby MF. The selection of sensors. characterization of a bell-tower by interferometric sensor. NDT&E Int
Prog Mater Sci 2001;46:461–504. 2007;40:390–6.