buildings

Article
Multiple Tests for Dynamic Identification of a Reinforced
Concrete Multi-Span Arch Bridge
Vincenzo Gattulli 1 , Francesco Potenza 2, * and Giulio Piccirillo 1

1 Department of Structural and Geotechnical Engineering, Sapienza University of Rome, Via Eudossiana 18,
00184 Rome, Italy; vincenzo.gattulli@uniroma1.it (V.G.); giulio.piccirillo90@gmail.com (G.P.)
2 Department of Engineering and Geology, University G. d’Annunzio of Chieti-Pescara, Viale Pindaro 42,
65127 Pescara, Italy
* Correspondence: francesco.potenza@unich.it

Abstract: This paper presents the results of an experimental dynamic campaign carried out on a
reinforced concrete multi-span arch bridge. Five expeditious ambient vibration tests were conducted
separately on five spans (one test in each span) of the bridge using only six piezoelectric uniaxial
accelerometers. Modal parameters were identified through the well-known Enhanced Frequency
Domain Decomposition (EFDD) procedure developed using Matlab R2021b software. At the same
time, a finite element model was accurately implemented through a commercial software (Midas
Civil) to evaluate the main modal features. A manual model update was successively pursued
varying the elastic modulus of the reinforced concrete to make the identified and numerical modes as
close as possible. A complete and suitable instrumentation to perform global experimental dynamic
tests is not always available. Recursive/Multiple tests have different advantages: handy, easily
executable, and could provide a more robust identification thanks to a statical characterization. The
paper aims to highlight the peculiarities of recursive/multiple dynamic tests on multi-span arch
bridges. The procedure also provides useful suggestions for designing a permanent and continuous
vibration-based monitoring system.
Citation: Gattulli, V.; Potenza, F.;
Piccirillo, G. Multiple Tests for
Keywords: multiple dynamic tests; multi-span arch bridges; modal identification; arch bridge
Dynamic Identification of a
modeling; manual model updating
Reinforced Concrete Multi-Span
Arch Bridge. Buildings 2022, 12, 833.
https://doi.org/10.3390/
buildings12060833
1. Introduction
Academic Editor: Alessandra Aprile
Over the past few years structural identification (SI) has become an intrinsic part of
Received: 17 March 2022 current civil engineering practice. This is due to the fact that accurate information coming
Accepted: 12 June 2022 from SI increases the level of knowledge of the structural behavior. SI usually is pursued
Published: 15 June 2022 processing vibrational measurements aiming to identify the main modal parameters: fre-
Publisher’s Note: MDPI stays neutral
quencies, shapes, and damping ratios. These parameters could be used to update a linear
with regard to jurisdictional claims in
and elastic finite element model in order to make more reliable the numerical forecast model.
published maps and institutional affil- Experimental Modal Analysis (EMA) [1] and Operational Model Analysis (OMA) [2] consti-
iations. tute two different methodologies to carry out SI. The former is an input–output technique
that requires the knowledge of the input (usually an artificial load to excite the structure)
and output (structural response) while the second is based only on the output measure.
In recent years, OMA techniques have had wider use because they do not need
Copyright: © 2022 by the authors. cumbersome instrumentation installed and, above all, the recording of measurements
Licensee MDPI, Basel, Switzerland. induced by ambient vibrations is sufficient (wind, traffic load, crowd). Among the various
This article is an open access article algorithms and procedures developed for performing SI by output-only data, Enhanced
distributed under the terms and Frequency Domain Decomposition and Stochastic Subspace Identification are the most
conditions of the Creative Commons known and applied [3,4].
Attribution (CC BY) license (https://
During the years, experimental dynamic and static analysis has increasingly been
creativecommons.org/licenses/by/
incorporated in civil engineering applications [5]. Moreover, among all different typologies
4.0/).

Buildings 2022, 12, 833. https://doi.org/10.3390/buildings12060833 https://www.mdpi.com/journal/buildings

 used. Consolidation of testing techniques, sensor and acquisition systems improvements,
development of efficient algorithms, and procedures for identification, large storage, and
data transmission capacities have made such types of tests easier and more implementa-
ble. The information extrapolated by the corresponding measurements aim to better un-
Buildings 2022, 12, 833 2 of 24
derstand the structural behavior, especially in the case of complex structures [8].
Bridges, as other civil engineering structures, are constantly exposed to various dy-
namic loads such as traffic or human-induced loads, time varying wind loads, and earth-
of non-destructive or minor-destructive tests [6,7], experimental dynamic tests are widely
quakes. To date, used.an increasingly
Consolidationconscious design ofsensor
of testing techniques, complex and slender
and acquisition systemsbridge struc-
improvements,
tures is not always sufficient of
development toefficient
avoid unexpected
algorithms, and and unpredictable
procedures vibrational
for identification, largephenom-
storage, and
ena, also highlighted during thecapacities
data transmission openinghave day of the
made suchLondon Millennium
types of tests footbridge
easier and more [9].
implementable.
The information extrapolated by the corresponding measurements
Such factors have raised interests in the dynamic tests of bridges, developing new tools aim to better understand
the structural behavior, especially in the case of complex structures [8].
for data processing, analysis, and especially for damage detection and modeling [10–12].
Bridges, as other civil engineering structures, are constantly exposed to various dynamic
Many testing methods loads suchandas algorithms in bridgeloads,
traffic or human-induced engineering
time varying have
wind been
loads,borrowed from To
and earthquakes.
mechanical engineering where dynamic
date, an increasingly conscious phenomena
design of complex and andexperimental modal is
slender bridge structures analysis
not always
sufficient to avoid unexpected and unpredictable vibrational
were investigated earlier. However, direct use of all these methods for bridge structures phenomena, also highlighted
encounters manyduring the opening day of the London Millennium footbridge [9]. Such factors have raised
problems related to the more complex nature of bridge materials (con-
interests in the dynamic tests of bridges, developing new tools for data processing, analysis,
crete, steel, brick,andcomposite
especially material)
for damageand structural
detection typologies
and modeling (beam-like,
[10–12]. Many testing suspended,
methods and
stay-cable). The selection of an appropriate testing method can be based
algorithms in bridge engineering have been borrowed from mechanical engineering on different con-
where
dynamic phenomena
siderations: (1) structural type and and size,experimental
(2) availablemodal analysis were(for
technologies investigated
example, earlier.
wiredHowever,
or
wireless), (3) strategic importance of the infrastructure, and (4) costs and effective applica- to
direct use of all these methods for bridge structures encounters many problems related
the more complex nature of bridge materials (concrete, steel, brick, composite material) and
bility. The last isstructural
often more decisive but, nevertheless, it could be overcome by recur-
typologies (beam-like, suspended, stay-cable). The selection of an appropriate test-
sive/multiple tests ingperformed
method can be inbased
different zonesconsiderations:
on different of the bridge. Indeed, type
(1) structural such andtypes of available
size, (2) tests
allow to know thetechnologies
actual dynamic behavior
(for example, wired especially
or wireless), in
(3) the cases
strategic of infrastructures
importance hav-
of the infrastructure,
and (4) costs and
ing a prevalent longitudinal effective applicability.
development. FigureThe last is often
1 shows more decisive
a simple scheme but,ofnevertheless,
the most it
could be overcome by recursive/multiple tests performed in different zones of the bridge.
popular dynamic tests on bridges where, as mentioned before, the ambient vibration ones
Indeed, such types of tests allow to know the actual dynamic behavior especially in the
are surely the most applied.
cases of infrastructures having a prevalent longitudinal development. Figure 1 shows a
simple scheme of the most popular dynamic tests on bridges where, as mentioned before,
the ambient vibration ones are surely the most applied.

BRIDGE
DYNAMIC
TESTS

AMBIENT FORCED FREE

VIBRATION VIBRATION VIBRATION
TESTS TESTS TESTS

RELEASING
SPECIAL WIND, DYNAMIC IMPACT DROPPING
TRAFFIC APPLIED
VEHICLES EARTHQUAKES EXCITERS HAMMER WEIGHT
DEFLECTION

Figure 1. Schematic view of the main dynamic tests on bridges.

Figure 1. Schematic view

In of
thethe main dynamic
beginning, tests on bridges.
the experimental identification of relevant dynamic properties of
bridges was carried out on the basis of conventional modal testing procedures previously
developed
In the beginning, theinexperimental
the fields of mechanical and aeronautical
identification engineering.
of relevant dynamic Suchproperties
tests involvedofthe
bridges was carried out on the basis of conventional modal testing procedures previouslyand
estimation of a set of Frequency Response Functions (FRFs) relating the applied force
corresponding response at several pairs of points along the structure. These required the use
developed in theoffields of mechanical and aeronautical engineering. Such tests involved
an articulated instrumentation chain needed for structural excitation, data acquisition, and
the estimation of signal
a set processing.
of Frequency Response
Although Functions
forced vibration tests(FRFs)
may leadrelating the applied
to very accurate force
modal estimates,
and corresponding theyresponse at several
present a strong pairswhen
drawback of points
dealingalong thestructures.
with large structure. Thesetherequired
Indeed, difficulty in
exciting the most significant modes of vibration in a low range
the use of an articulated instrumentation chain needed for structural excitation, data of frequencies with sufficient
ac-
energy is often very demanding in terms of both costs and specialized technical personnel.
quisition, and signal processing. Although forced vibration tests may lead to very accurate
For example, in the case of small and medium-size structures, the excitation can be induced
modal estimates, bythey presenthammer
an impulse a strongthat,drawback
thanks to itswhen
capacitydealing witha large
of providing structures.
wide-band In- to
input, is able
deed, the difficulty in exciting the most significant modes of vibration in a low
excite different modes of vibration. For this reason, the main drawbacks of these types range of of
frequencies with tests concern the
sufficient lack ofisenergy
energy oftento very
supplydemanding
the whole experimental
in termslayout.
of bothTherefore,
costs ambient
and
 the main drawbacks of these types of tests concern the lack of energy to supply the whole
experimental layout. Therefore, ambient vibration tests gradually became the most ap-
plied, especially for evaluating serviceability conditions or the level of structural rehabil-
itation
Buildings 2022, 12, 833 [13–17]. 3 of 24
The results obtained from the application of different output-only modal identifica-
tion techniques applied to ambient response data collected during two dynamic tests of a
cable-stayed bridge and tests
vibration the subsequent
gradually became finite
theelement model
most applied, updating
especially are presented
for evaluating in
serviceability
conditions or the level of structural rehabilitation [13–17].
[18]. The first test, performed by using a traditional data acquisition system with servo-
The results obtained from the application of different output-only modal identification
accelerometers, was aimed at investigating the vertical dynamic characteristics of the
techniques applied to ambient response data collected during two dynamic tests of a
bridge. In the second test, anbridge
cable-stayed innovative
and theradar vibrometer
subsequent was used
finite element for updating
model non-contact meas-
are presented
urement of deflectionin [18].time
Theseries of the
first test, forestays
performed aiming
by using at identification
a traditional of localsystem
data acquisition natural with
frequencies of theservo-accelerometers,
stay cables. In thewas aimed at investigating
numerical the vertical
analysis, vibration dynamic
modes werecharacteristics
evaluated of
the bridge. In the second test, an innovative radar vibrometer was used for non-contact
using a 3D finite element model of the bridge and the information obtained from the field
measurement of deflection time series of the forestays aiming at identification of local
tests, combined with natural a frequencies
classic systemof the identification
stay cables. In the technique,
numerical provided a linear
analysis, vibration elastic
modes were
model, accuratelyevaluated
fitting the
using modal parameters
a 3D finite of theofbridge
element model in its
the bridge andactual condition.obtained
the information An-
from the field
other structural typology hastests,
beencombined
analyzed withina[19].
classic
Insystem identification
this case, the SI has technique, provided a
been strongly
linear elastic model, accurately fitting the modal parameters
affected by the mechanical properties of the bearing devices supporting the deck super- of the bridge in its actual
condition. Another structural typology has been analyzed in [19]. In this case, the SI has
structures. Subsequently, the identified features have been compared with those pre-
been strongly affected by the mechanical properties of the bearing devices supporting the
dicted by a preliminary finite element
deck superstructures. model. This
Subsequently, latter has
the identified beenhave
features takenbeeninto account
compared withasthose
a
base line model for monitoring
predicted and condition
by a preliminary assessment
finite element programs
model. This related
latter has to this
been taken intoinfra-
account
structure. as a base line model for monitoring and condition assessment programs related to this
infrastructure.
Historically, arch bridges have been mostly known as masonry structures (Figure
Historically, arch bridges have been mostly known as masonry structures
2a,b [20,21]) but (Figure
worldwide examples
2a,b [20,21]) have been
but worldwide also have
examples realized using
been also different
realized materials
using different mate-
(Figure 2c,d [22,23]).
rials (Figure 2c,d [22,23]).

Figure 2. ExamplesFigure
of arch bridges:of(a)
2. Examples Hacikiri
arch bridges: bridge [20];bridge
(a) Hacikiri (b) Villanova bridgebridge
[20]; (b) Villanova [21]; [21];
(c) Viaduct
(c) Viaduct
Cairate [22]; (d) S. Giovanni
Cairate [22];Paolo II bridgePaolo
(d) S. Giovanni [23].II bridge [23].

In [20], a model calibration has been pursued based on the information coming from
In [20], a model calibration has been pursued based on the information coming from
both dynamic and static tests. Afterwards, refined models, representative of the experi-
both dynamic and static
mental tests. Afterwards,
behavior, have been usedrefined models,
for structural representative
assessment especiallyof
in the experi-
the verification
mental behavior,ofhave been usedtofor
the capabilities structural
withstand trainassessment especially
loads. The results in the
of ambient verification
vibration of
tests applied
 ing the asymmetry configuration of the viaduct, vertical and lateral modes have been suc-
cessfully identified processing output-only measurements by two classical procedures
such as Peak Picking and Enhanced Frequency Domain Decomposition. An excellent
agreement
Buildings 2022, 12, 833 has been found between the two sets. Further, manual model updating has 4 of 24
been pursued varying the Young’s moduli of the main concrete components. Recently,
ambient vibration tests have been carried out also on modern concrete arch bridges built
using both reinforced concreteconcrete
on reinforced and steel.archAn example
bridges can be
have been found in
illustrated in [23], where
[22]. In dynamic
this work, notwith-
standing the asymmetry configuration of the viaduct, vertical
tests have been performed with different loading conditions. Even in this case, suitable and lateral modes have been
successfully identified processing output-only measurements by two classical procedures
numerical models have been developed to interpret and represent experimental evidence.
such as Peak Picking and Enhanced Frequency Domain Decomposition. An excellent
In [24], the resultsagreement
of a hugehas experimental
been found betweencampaignthe two(845 dynamic
sets. Further, load
manual tests)
modelcarried
updatingouthas
on 11 masonry arch beenbridges
pursuedto characterize
varying the Young’sthe moduli
verticalofand lateral
the main response
concrete by different
components. Recently,
parameters such as ambient
train vibration
formation testsand
have been carried
speed, span out also on
length, modern concrete
rise/span ratio, andarchfirst
bridges
nat- built
using both reinforced concrete and steel. An example can be found in [23], where dynamic
ural frequencies have been highlighted.
tests have been performed with different loading conditions. Even in this case, suitable
Here, in the illustrated
numerical models examples the developed
have been modal identification
to interpret and is represent
based on the processing
experimental evidence.
of a few numbers In of[24],
experimental
the results oftests.
a huge Aexperimental
minimum statistical
campaign (845 in-depth
dynamic analysis is carried
load tests) carried out
out only processing on 11themasonry
same measurements through different
arch bridges to characterize the verticalidentification
and lateral response procedures.
by different
parameters such as train formation and speed, span
Moreover, as mentioned, a complete investigation of large infrastructures becomes length, rise/span ratio, and first natural
diffi-
frequencies have been highlighted.
cult using both wired Here, and wireless equipment. Indeed, in the first case there are problems
in the illustrated examples the modal identification is based on the processing of
related to the deployment
a few numbers of electric/communication
of experimental tests. A minimum cables statistical
while inin-depth
the second oneis itcarried
analysis is
possible to encounter issues
out only of synchronization.
processing the same measurementsFor these reasons,
through in this
different work a procedures.
identification way to
overcome such obstacles will be shown through multiple experimental tests realizeddifficult
Moreover, as mentioned, a complete investigation of large infrastructures becomes in
using both wired and wireless equipment. Indeed, in the first case there are problems
different portions of the infrastructure. The first part will be dedicated to the results of
related to the deployment of electric/communication cables while in the second one it is
experimental researchpossiblerelated to the issues
to encounter dynamic behavior of a For
of synchronization. reinforced concrete
these reasons, multi-span
in this work a way to
arch bridge. In the second one, critical aspects in the SI and manual model
overcome such obstacles will be shown through multiple experimental tests realized updating of in
such complex structure
differentwill be pointed
portions out. Moreover,
of the infrastructure. the part
The first results
will provided
be dedicated bytostatistical
the results of
experimental research related to the dynamic behavior
processing of the information coming from recursive or multiple experimental dynamic of a reinforced concrete multi-span
arch bridge. In the second one, critical aspects in the SI and manual model updating of
tests will be illustrated. The information acquired in the fast experimental campaign fur-
such complex structure will be pointed out. Moreover, the results provided by statistical
nishes relevant suggestions
processing offor the designcoming
the information of a permanent
from recursive structural
or multiple health monitoring
experimental dynamic
installation basedtests
on vibration measurements.
will be illustrated. The information acquired in the fast experimental campaign
furnishes relevant suggestions for the design of a permanent structural health monitoring
2. Dynamic Testsinstallation
on the Case based on vibration measurements.
Study
Since the dawn of time,
2. Dynamic theoncity
Tests of Rome
the Case Study has been crossed by the “Tevere” river.
Thanks to its presence,Since
a wide number
the dawn of bridges
of time, the city ofwere
Rome constructed
has been crossed tobycross the same
the “Tevere” river.
river. Thanks
to its presence, a wide number of bridges were constructed to cross the
In this context, one of these bridges is the subject matter of this research and it is called same river. In this
context, one of these bridges is the subject matter of this research and it is called “Ponte della
“Ponte della Magliana”. An overview and zoomed-in view from above of its location are
Magliana”. An overview and zoomed-in view from above of its location are reported in
reported in FigureFigure
3. 3.

Figure 3. OverviewFigure
and zoom of theand
3. Overview “Ponte
zoomdella Magliana”
of the “Ponte dellabridge in bridge
Magliana” the urban
in thecontext.
urban context.

The bridge connects the “Portuense” and “Ostiense” districts of Rome, respectively,
located in northwest and southeast directions with respect to the river. It was designed in
1930 and built to be the entrance to the western side of the new “EUR” district. Its
R PEER REVIEW 5 of 23
Buildings 2022, 12, 833 5 of 24

complete realization was only after the second world war. It is currently part of a larger
The bridge connects the “Portuense” and “Ostiense” districts of Rome, respectively,
highway that, in the westindirection,
located northwest andcontinues
southeasttowards the “Portuensi”
directions with hillsItand
respect to the river. was the high-
designed
way to “Fiumicino”inairport
1930 andwhile
built toonbe the east it to
the entrance leads towards
the western sidethe “EUR”,
of the new “EUR”“Tredistrict.
Fontane”,
Its
complete realization was only after the second world war. It is currently part of a larger
and “Laurentina” districts.
highway that, in the west direction, continues towards the “Portuensi” hills and the highway
The structure tois “Fiumicino”
a reinforced concrete
airport while on arch bridge,
the east it leads with
towards a the
length
“EUR”,of “Tre
about 225 m.
Fontane”, and It
develops linearly through 7 spans
“Laurentina” districts.and rests on 2 abutments and 5 piers, 2 of which are in
The structure is a reinforced concrete arch bridge, with a length of about 225 m. It
the riverbed and 4 outside the riverbed. The length of each span is about 30 m except for
develops linearly through 7 spans and rests on 2 abutments and 5 piers, 2 of which are in
the central and moveable span
the riverbed andthat presents
4 outside a length
the riverbed. Theof about
length 40 m.
of each spanThe static
is about scheme
30 m except foris a
supported beam, asthe shown
central by
andthe original
moveable span drawings
that presentspreserved by the
a length of about municipal
40 m. technical
The static scheme is a
supported beam, as shown by the original drawings preserved by the municipal technical
office [25] and illustrated in Figure 4. Such configuration aimed to minimize the internal
office [25] and illustrated in Figure 4. Such configuration aimed to minimize the internal
stresses due to possible
stressesfailures related
due to possible to the
failures supports
related of the ofisostatic
to the supports system.
the isostatic system.

Figure 4. Original static scheme

Figure of the
4. Original staticbridge
scheme (found in [25]).
of the bridge (found in [25]).

This static scheme divides the bridge into four isostatic substructures that are capa-
This static scheme divides the bridge into four isostatic substructures that are capable
ble of absorbing possible structural failures (e.g., subsidence) without generating states
of absorbing possible structural
of internal failures
compulsion. (e.g.,
Such subsidence)
scheme withouttogenerating
allows the structure remain in thestates
elastic of in-
field
following small altimetric and planimetric variations of piers
ternal compulsion. Such scheme allows the structure to remain in the elastic field follow- and abutments and so, for
this reason, is particularly suitable for river crossings. The fundamental structural element
ing small altimetricthatand planimetric variations of piers and abutments and so, for this
allows such configuration is the Gerber saddle that represents a peculiarity of this
reason, is particularly suitable
bridge. for river
As visible crossings.
in Figure The composition
5a, the overall fundamental showsstructural element
four arch-shaped that
spans
linked together
allows such configuration is theby three Gerber
Gerber saddle saddles. The concrete adecks
that represents have a multi-box
peculiarity of this section,
bridge.
separated by five septa with a thickness of 0.50 m. Since each span presents a longitudinal
As visible in Figurecurvature,
5a, the the overall composition shows four arch-shaped spans linked to-
height of such transversal section is variable along the span length with a
gether by three Gerber
maximum saddles. The concrete
and minimum decks have
value corresponding a multi-box
to the supports andsection, separated
central line, by
respectively.
five septa with a thickness of 0.50 m. Since each span presents a longitudinal curvature,
On the extrados of the section, there is a concrete slab, with an average thickness of 0.20 m,
the height of such transversal section is variable along the span length with a maximum
and minimum value corresponding to the supports and central line, respectively. On the
extrados of the section, there is a concrete slab, with an average thickness of 0.20 m, while
 Buildings 2022, 12, 833 6 of 24

while in the intrados there is a slab showing a variable thickness only in particular zones of
the bridge.
The steel deck (that was previously moveable) has a static layout of a supported beam.
The main beams are made up of pairs of double T beams variable in height and thickness.
Between such two beams, double T sections are positioned transversely. The extrados slab,
R PEER REVIEW in reinforced concrete and connected by pegs to the steel beams, is 0.20 m thick and6restsof 23
on UPN200 steel profiles. In Figure 5b,c, two examples of transversal sections are reported
with an open profile and multi-connected, respectively.

Figure 5. (a) Half left (above) and half right (below) lateral section of the bridge. Highlights of the
Figure 5. (a) Half leftspan
(above) and half right (below) lateral section of the bridge. Highlights of the
typologies: alternation between arch-shaped and Gerber saddles. (b,c) Two types of transversal
span typologies: alternation between
sections: (a) arch-shaped
open profile and Gerber saddles. (b,c) Two types of transver-
and (b) multi-connected.
sal sections: (a) open profile and (b) multi-connected.
2.1. Experimental Layouts
The dynamic tests were carried out in a one-day experimental campaign through
2.1. Experimental Layouts
which accelerometric measurements were recorded. The tests were implemented during
The dynamic atests very sunny
were day (30 July
carried 2020)
out in and the traffic experimental
a one-day conditions were very intense. Ambient
campaign through
vibration tests were conducted on the five central spans of the bridge using a 12-channel
which accelerometric measurements were recorded. The tests were implemented during
data acquisition system, LMS SCADAS XS, with n.6 piezoelectric uniaxial accelerometers,
a very sunny day PCB (30 model
July 2020)
393B31,andand athe traffic
Smart Scope. conditions were very intense. Ambient
LMS SCADAS XS constitutes
vibration tests were conducted on the five central the main part ofof
spans thethe
databridge
acquisition system
using and, thanks
a 12-channel
to its manageability, it is very suitable to optimize the implementation of the dynamic tests.
data acquisition system, LMS SCADAS XS, with n.6 piezoelectric uniaxial accelerometers,
It is very useful for carrying out both real and reduced-scale tests. The main characteristics
PCB model 393B31,toand a Smartare
be underlined Scope.
the following:
LMS SCADAS1.XS constitutes the
The board has a very main
smallpart of the(asdata
dimension acquisition
a tablet system
pc) and, moreover, and,
it has thanks
a built-in
to its manageability, itbattery.
is very suitable to optimize the implementation of the dynamic
tests. It is very useful for carrying out both real and reduced-scale tests. The main charac-
teristics to be underlined are the following:
1. The board has a very small dimension (as a tablet pc) and, moreover, it has a built-in
Buildings 2022, 12, 833 7 of 24

2. It can have three different modes of operation: Wi-Fi (connected to Smart Scope),
stand-alone, and front-end.
3. It can support 12 analog channels.
The Smart Scope is a tablet on which the user can set up, control, and manage the
measurement template and carry out online data processing. The most relevant parameters
to be set up are sensor name, point ID and point direction, typology of the physical
quantity to be recorded and its unit of measure, sensitivity, and acquisition sample rate.
The piezoelectric accelerometers, employed in the dynamic tests, are PCB 393B31 realized
with ICP technology that requires only an inexpensive, constant-current signal conditioner
to operate. Their main features are illustrated in the following Table 1.

Table 1. Characteristics of the piezoelectric accelerometers PCB 393B31.

Feature Value Feature Value

Frequency range
Sensitivity (±5%) 10.0 V/g 0.1–200 Hz
(±5%)

Broadband resolution 0.000001 g rms Transverse sensitivity ≤5%

Measurement range 0.5 g pk Resonant frequency ≥700 Hz

Overload limit
Non-linearity ≤1% ±40 g pk
(shock)

The link between the sensors and acquisition system is completed by a complementary
instrumentation. This latter is composed of different typologies of coaxial cables used
for both transmissions of the measured data and power supply. In particular, two cables
(coaxial cable RG 178/179 and custom cable 052BR010AC), from a side, have to be linked to
the board LMS SCADAS and to the piezoelectric sensor, respectively, while the third cable
(coaxial cable RG58) must be linked in the other free end of the two previous cables (it only
has an extension function).
Moreover, in order to realize a reliable link between the sensor and structure, a
customized aluminum cube with a central thread for screw insertion was realized, in each
face, needed for the connection of the accelerometric sensor. Furthermore, in one of the six
cube faces, four magnetic elements were introduced for a rapid and easy connection in steel
or iron structure points. In the photos reported in Figure 6, the presence of a bicycle lane is
visible, located in both external lines of the bridge, that allowed an easy implementation
and execution of the dynamic tests.
The five spans, the object of the dynamic tests, are indicated in Figure 7 where,
moreover, a well-defined code was assigned at each sensor position for an unequivocal
identification. In each position (highlighted by green circles in Figure 7), two piezoelectric
uniaxial accelerometers were installed in a biaxial configuration in order to identify the
dynamic behavior of the following directions:
• Y: horizontal and transversal direction with respect to the longitudinal axis of the
bridge.
• Z: vertical direction with respect to the longitudinal axis of the bridge.
Three sensor positions were designed for each span. One position is always located in
the middle of the span length and so, in the case of the Gerber saddle span, it is in the middle
of the corresponding saddle. The other two positions were selected on the right and left with
respect to the middle with the aim to also identify anti-symmetric modes. Moreover, in each
position, two PCB accelerometers were installed: in the Z- and Y-directions, respectively. For
each position, the following identification parameters were assigned:
• Position on the span: R: right; M: midpoint; L: left.
• Number of the span: 1, 2, 3, 4, and 5.
• Direction of the measurement: Z, Y.
 distance from such midpoint. Regarding the two sensors located in the midpoint, unfor-
tunately, for technical and safety reasons, it was impossible to realize a reliable connection
Buildings 2022, 12, 833
to the structure; therefore, their recorded accelerations were not considered in the subse-8 of 24
quent processing (this position is highlighted with a red circle in Figure 7).

Figure 6. Accelerometers and examples of the experimental layout installation on the bridge.
Figure 6. Accelerometers and examples of the experimental layout installation on the bridge.
22, 12, x FOR PEER2022,
Buildings REVIEW
12, 833 9 of 23 9 of 24

Figure 7. Planimetric
Figure(above) and lateral
7. Planimetric (below)
(above) deployment
and lateral of the five experimental
(below) deployment layouts. layouts.
of the five experimental The The
circles representcircles
the location of the operating (green) and not operating (red) sensors. R: right; M:
represent the location of the operating (green) and not operating (red) sensors. R: right;
midpoint; L: left.M: midpoint; L: left.

2.2. Modal Identification: General

The whole timeBackground
of acquisition was not the same for each span. In particular, 1800 s was
adopted for the span n◦ 3 while 1200 s for the other spans. The sample frequency for data
One of the more traditional approaches applied to◦ estimate the modal parameters of
acquisition was 200 Hz for all tests. Spans n 1 and n◦ 5 are characterized by the presence
a structure is the Peak Picking (PP) method [26]. Such method could lead to reliable results
of two Gerber saddles (one for each span) realized by a simply supported deck that divides
if, as basic assumptions, the modes
the corresponding spanto into
be identified have low
three elements. damping
In order andthe
to study aredifferent
well-sepa-structural
rated from each other. In
response of fact, when
the three a lightly
elements, damped
one structure
accelerometer wasisset
subjected to a random
on the midpoint deck and the
excitation, the output
other two auto-spectral
were positioneddensities at any
externally response
to the saddle.point n◦ 2reach
Spanswill and na◦ maximum
4, differently from
◦ ◦
at frequencies related to the modal signatures of the structures. Since narrow-band
spans n 1 and n 5, did not have Gerber saddles but an arched profile. In peaks
this case, one
in the frequency accelerometer was set onof
response function thelightly
midpoint and themechanical
damped other two were locatedoccur
systems at a certain
at thedistance
frequencies corresponding to normal system modes (resonance frequencies), peaks inofthe
from the midpoint one. A particular configuration can be observed in the case span n◦ 3.
Indeed, it shows a central steel saddle Gerber. Originally, this latter was a removable steel
auto-spectral densities and cross-spectral densities can be generally assumed to represent
element span but, over time, it was modified and fixed. Moreover, this element presents
the normal modes of the structure.
on one end (side EUR direction, left in the Figure 7) a simple constraint (like a roller),
An evolution
whileofinthethePP method
other one itisisthe Frequency
rigidly connected Domani
to twoDecomposition (FDD).Additionally,
cantilever steel beams. This
technique involves
in thisthe following
case, mainwas
one position steps:
set on the midpoint and the other two at a certain distance
1. Estimate offrom
thesuch midpoint.
spectral matrix.Regarding the two sensors located in the midpoint, unfortunately,
2. for technical
Singular Value and safety(SVD)
Decomposition reasons, it was
[27] impossible
of the spectral to realize
matrix ataeach
reliable connection to the
frequency.
structure; therefore, their recorded accelerations were not considered in the subsequent
3. Inspection of the curves representing the singular values to identify the resonant fre-
processing (this position is highlighted with a red circle in Figure 7).
quencies and estimate the corresponding modal shape using the information con-
tained in the singular vectors of the SVD.
FDD is a rather simple procedure and represents a significant improvement in the PP
since:
• SVD is an effective method to analyze, in depth, the spectral matrix characteristics
and the evaluation of the mode shapes is automatic and significantly easier than in
the PP.
Buildings 2022, 12, 833 10 of 24

2.2. Modal Identification: General Background

One of the more traditional approaches applied to estimate the modal parameters
of a structure is the Peak Picking (PP) method [26]. Such method could lead to reliable
results if, as basic assumptions, the modes to be identified have low damping and are
well-separated from each other. In fact, when a lightly damped structure is subjected to a
random excitation, the output auto-spectral densities at any response point will reach a
maximum at frequencies related to the modal signatures of the structures. Since narrow-
band peaks in the frequency response function of lightly damped mechanical systems
occur at the frequencies corresponding to normal system modes (resonance frequencies),
peaks in the auto-spectral densities and cross-spectral densities can be generally assumed
to represent the normal modes of the structure.
An evolution of the PP method is the Frequency Domani Decomposition (FDD). This
technique involves the following main steps:
1. Estimate of the spectral matrix.
2. Singular Value Decomposition (SVD) [27] of the spectral matrix at each frequency.
3. Inspection of the curves representing the singular values to identify the resonant
frequencies and estimate the corresponding modal shape using the information con-
tained in the singular vectors of the SVD.
FDD is a rather simple procedure and represents a significant improvement in the PP
since:
• SVD is an effective method to analyze, in depth, the spectral matrix characteristics and
the evaluation of the mode shapes is automatic and significantly easier than in the PP.
• The FDD technique is able to detect closely spaced modes. In such instances, more
than a singular value will reach a maximum in the neighborhoods of a given frequency
and every singular vector, corresponding to a non-zero singular value, is a mode shape
estimate.
• Damping ratios can be identified through a refinement of the FDD technique, called
Enhanced Frequency Domain Decomposition (EFDD).
The EFDD technique is based on the fact that the first singular value in the neigh-
borhoods of a resonant peak is the auto-spectral density of a modal coordinate. Hence,
taking back the partially identified auto-spectral density of the modal coordinate in the time
domain, by inverse fast Fourier transform, a free decay time domain function is obtained.
This latter represents the autocorrelation function of the modal coordinate. The natural
frequencies and the related damping ratios are thus simply found using the peak of the
auto-spectral densities and logarithmic decrements, respectively.

2.3. Modal Identification: Case Study

The identification of modal parameters was developed using two different output-only
procedures: the Peak Picking (PP) and the Enhanced Frequency Domain Decomposition
(EFDD). The EFDD procedure is a refinement of the Frequency Domain Decomposition
technique. Both the PP and FDD/EFDD methods are based on the evaluation of the spectral
matrix (i.e., the matrix of cross-spectral densities) in the frequency domain. In Figure 8,
some examples of the acquired acceleration during the tests are reported. In particular, they
were recorded by the experimental layout installed in span 3.
It can be observed that the accelerations related to the sensors of the right position
(Figure 8a,b) show a lower amplitude with respect to the left ones (Figure 8c,d). Addition-
ally, the corresponding standard deviations (σ) show low values: σR3Z = 1.1286 × 10−5 g,
σR3Y = 9.7175 × 10−6 g, σL3Z = 1.8621 × 10−5 g, and σL3Y = 9.6748 × 10−6 g. They are
lower in the right position and in general the transversal direction (Y) is harder to excite.
Moreover, a wide presence of spikes due to the passage of trucks of large dimension are
clearly visible.
 The identification of modal parameters was developed using two different output-
only procedures: the Peak Picking (PP) and the Enhanced Frequency Domain Decompo-
sition (EFDD). The EFDD procedure is a refinement of the Frequency Domain Decompo-
sition technique. Both the PP and FDD/EFDD methods are based on the evaluation of the
spectral matrix (i.e., the matrix of cross-spectral densities) in the frequency domain. In
Buildings 2022, 12, 833 11 of 24
Figure 8, some examples of the acquired acceleration during the tests are reported. In par-
ticular, they were recorded by the experimental layout installed in span 3.

Acceleration R3Z Acceleration R3Y

0.0002 0.0002
a[g] a[g]
0 0

(a) (b)
−-0.0002 −-0.0002

0 300 600 900 1200 1500 1800 0 300 600 900 1200 1500 1800
t[s] t[s]
Acceleration L3Z Acceleration L3Y
0.0008 0.0008
a[g] a[g]
0 0

(c) (d)
−-0.0008 −-0.0008

0 300 600 900 1200 1500 1800 0 300 600 900 1200 1500 1800
t[s] t[s]

Figure Examplesofof
8. Examples
Figure 8. accelerometric
accelerometric measurements
measurements recorded
recorded in span
in the the span
n3 in n3
thein the following
following corre-
corresponding positions
sponding positions and directions:
and directions: (a)
(a) R3Z R3ZR,3Y(b)
, (b) R3Y
, (c) , (c)
L3Z L3Z(d)
, and , and
L3Y.(d) L3Y .

For thebecase
It can studied,
observed thethe
that modal identification
accelerations related was
to performed
the sensorsthrough a procedure
of the right position
implemented in Matlab software found in the literature [28]. First of all,
(Figure 8a,b) show a lower amplitude with respect to the left ones (Figure 8c,d). Addition- the accelerometric
signals
ally, thewere subjected to
corresponding a lowpass
standard filter with
deviations (σ)ashow
frequency range between
low values: 0 and
σR3Z = 1.1286 20
10−5Hz.
Buildings 2022, 12, x FOR PEER REVIEW 11 of× 23 g,
The Power Spectral Densities (PSDs) were obtained with the periodogram function while
σR3Y = 9.7175 × 10−6 g, σL3Z = 1.8621 × 10−5 g, and σL3Y = 9.6748 × 10−6 g. They are lower in the
mode shapes with the Automated Frequency Domain Decomposition (AFDD) function [28].
right position and in general the transversal direction (Y) is harder to excite. Moreover, a
In Figure 9 the PSDs in Y and Z directions calculated by processing the measurements
wideInpresence of spikes due to the passage of trucks ofby large dimension are clearly visible.
[28].
obtainedFigure 9 the
during PSDs
the in
test Y and
on theZspan n◦ 3 calculated
directions are shown. Lookingprocessing atthe
themeasurements
results provided
For
obtained the case
duringitthe studied, the
test on the modal identification was performed through a procedure
by such PSDs, is possible tospan n° 3 that
observe are shown. Looking at the
in the Z-direction, results provided
well-defined peak by values
implemented
such PSDs, it isinpossible
Matlabto software
observe found
that ininthe
theZ-direction,
literature [28]. First of all,
well-defined thevalues
peak accelerometric
cor- in
corresponding to the structural modes are difficult to recognize (Figure 9a,c). Instead,
signals were
responding to subjected
the to a lowpass
structural arefilter withto arecognize
frequency range9a,c).
between 0 and 20 Hz.
the Y-direction peak valuesmodes
are clear difficult
and easily identifiable (Figure
(Figure Instead,
9b,d). in the
By comparing
The Power
Y-direction Spectral Densities (PSDs) were obtained with the periodogram function thewhile
the results peak
foundvalues
in theare clearcalculated
PSDs and easilyfor identifiable (Figure 9b,d).
both directions, By comparing
the frequencies reasonably
mode
results shapes
found with the Automated Frequency Domain Decomposition (AFDD) function
associated tointhe
thestructural
PSDs calculated
modesfor bothidentified.
were directions, They
the frequencies
are reported reasonably asso-
in the following
ciated to
Table 2. the structural modes were identified. They are reported in the following Table 2.

2 × 10-11 1.5 × 10-11

15.47
R3Z R3Y 19.53
A [g2/Hz]
A [g2/Hz]

9.77

14.40
17.31 12.50
10.44
9.77 19.53
(a) (b)

f [Hz] f [Hz]
-11 -12
1 × 10 3 × 10
12.50 L3Y 12.50
L3Z
A [g2/Hz]

A [g2/Hz]

9.77

19.53
9.77 19.53
16.68

(c) (d)

f [Hz] f [Hz]

PSDsrelated
Figure9.9.PSDs
Figure relatedtotothe
thetime
timehistories
historiesillustrated
illustratedininFigure
Figure8:8:(a)(a)R3Z
R,3Z(b)
, (b)
R3YR, 3Y
(c), (c)
L3Z,Land
3Z , and
(d) (d) L3Y .
L3Y.

Table 2. Peak values shown by the PSDs of Figure 9 associable to structural modes.

Mode Frequency (Hz)

1 9.77
Direction Z
 Buildings 2022, 12, 833 12 of 24

Table 2. Peak values shown by the PSDs of Figure 9 associable to structural modes.

Mode Frequency (Hz)

1 9.77
Direction Z
2 19.53

1 9.77
Direction Y 2 12.50
3 19.53

Due to the fact that in the Z-direction it is hard to define peak values (this is also
a consequence of the very small amplitude measured), a probabilistic approach was im-
plemented on all values (in all five tests) obtained in both directions for a more robust
identification. The probabilistic approach was organized based on the following three
phases:
1. Identification of the presumable frequencies coming from the measurements obtained
in the five dynamic tests carried out (in total 28 accelerations recorded, 14 in each
direction Y and Z).
FOR PEER REVIEW 12 of 23
2. Identification of the frequencies with the highest number of appearances.
3. Graphical representation with a bar histogram for a better visualization.
The results of the probabilistic approach are illustrated in Figure 10. In such graphs, the
frequencies most detected by analyzing the accelerations recorded in both Z- (Figure 10a)
18.18 Hz only inand
theY- Z-direction and 12.50
(Figure 10b) directions in allHz only tests
dynamic in the Y-direction,
are reported. appearthe
Substantially, more
two fre-
quently than others. The mostly
frequencies latterfound
one incould be morearereliable
both directions 9.77 and since it Two
19.53 Hz. has other
beenfrequencies,
found more
18.18 Hz only in the Z-direction and 12.50 Hz only in the Y-direction, appear
times with respect to the first one. The final set of the identified frequencies is shown in more frequently
than others. The latter one could be more reliable since it has been found more times with
Table 3. respect to the first one. The final set of the identified frequencies is shown in Table 3.

Z-direction Y-direction (19.53; 11)

(9.77; 9) (12.50; 9)
(9.77; 9)
appearances

appearances
Number of

Number of

(19.53; 5)
(18.18; 4)

(a) (b)

f [Hz] f [Hz]

Figure 10. Frequencies most detected processing the measurements in Z- (a) and Y- (b) directions in
Figure 10. Frequencies most detected processing the measurements in Z- (a) and Y- (b) directions in
all dynamic tests. Within round brackets: (frequency identified, number of appearances).
all dynamic tests. Within round brackets: (frequency identified, number of appearances).
Table 3. Identified experimental frequencies in Z- and Y-directions.
Table 3. Identified experimental frequencies in Z- and Y-directions.
Mode Frequency (Hz)
Mode 1 Frequency
9.77 (Hz)
Direction Z 2 18.18
1 9.77
3 19.53
Direction Z 2 18.18
1 9.77
Direction Y
3 2
19.53
12.50
1 3 9.77
19.53
Direction Y 2 12.50
3 19.53

Regarding the experimental layout, due to the presence of intensive traffic it was not
Buildings 2022, 12, 833 13 of 24

Regarding the experimental layout, due to the presence of intensive traffic it was
not possible to implement a configuration in the center longitudinal line of the roadway.
Of course, if such measurements were recorded they would have helped in the correct
identification of the coupled modes.
With the aim to deepen the statistical analysis, in Tables 4 and 5 the number of appear-
ances of the first three identified frequencies with varying numbers of the experimental tests
are reported. In particular, in Table 4 the trend for the modes in the Z-direction is illustrated
while in Table 5 for those in the Y-direction. It is evident that such frequencies are not
always identifiable for different motivation: (1) they could be referred to local modes, (2) in
correspondence with a zero-modal coordinate, or (3) due to a very low amplitude (that
means a very low signal/noise ratio). For these reasons, in Figure 11 the dependence of the
probability of appearance with varying numbers the tests is shown. Some observations are
the following:
1. The modes corresponding to the frequencies 9.77 Hz (in both directions) and 12.50 Hz
appear in almost all PSDs showing the same final probability of 64.29%. They could
be associated to global modes.
2. A little doubt could be raised by the mode at 19.53 Hz. Indeed, such frequency is not
always visible in the Z-direction (35.71%) but, regardless, it is commonly found in the
Y-direction (78.57%).
3. The mode at 18.18 Hz is the most difficult to understand. Substantially, it is visible
only in the Z-direction for the first two spans (28.57%). Some other suggestions could
be provided by looking at the modes of the numerical model.

Table 4. Number of appearances varying the number of tests: Z-direction.

Z-Direction
Test/Spam f1 (9.77 Hz) f2 (18.18 Hz) f3 (19.53 Hz)
N. App/N. PSD N. App/N. PSD N. App/N. PSD
1 1/3 (R1Z ) 2/3 (M1Z , R1Z ) 0/3 (-)

2 3/6 (L2Z , R2Z ) 4/6 (L2Z , M2Z ) 1/6 (M2Z )

3 5/8 (L3Z , R3Z ) 4/8 (-) 3/8 (L3Z , R3Z )

4 7/11 (M4Z , R4Z ) 4/11 (-) 3/11 (-)

5 9/14 (M5Z , R5Z ) 4/14 (-) 5/14 (M5Z , R5Z )

Table 5. Number of appearances varying the number of tests: Y-direction.

Y-Direction
Test/Spam f1 (9.77 Hz) f2 (12.50 Hz) f3 (19.53 Hz)
N. App/N. PSD N. App/N. PSD N. App/N. PSD
1 1/3 (R1Y ) 1/3 (L1Y ) 3/3 (L1Y , M1Y , R1Y )

2 3/6 (L2Y , R2Y ) 3/6 (L2Y , R2Y ) 5/6 (M2Y , R2Y )

3 5/8 (L3Y , R2Y ) 5/8 (L3Y , R3Y ) 7/8 (L3Y , R3Y )

4 6/11 (R4Y ) 7/11 (L4Y , R4Y ) 8/11 (R4Y )

5 9/14 (L5Y , M5Y , R5Y ) 9/14 (L5Y , R5Y ) 11/14 (L5Y , M5Y , R5Y )
 N. App/N. PSD N. App/N. PSD N. App/N. PSD
1 1/3 (R1Y) 1/3 (L1Y) 3/3 (L1Y, M1Y, R1Y)
2 3/6 (L2Y, R2Y) 3/6 (L2Y, R2Y) 5/6 (M2Y, R2Y)
3 5/8 (L3Y, R2Y) 5/8 (L3Y, R3Y) 7/8 (L3Y, R3Y)
Buildings 2022, 12, 833 4 6/11 (R4Y) 7/11 (L4Y, R4Y) 8/11 (R4Y) 14 of 24
5 9/14 (L5Y, M5Y, R5Y) 9/14 (L5Y, R5Y) 11/14 (L5Y, M5Y, R5Y)

Direction Z Direction Y
100
f1Z (9.77 Hz) f2Z (18.18 Hz)
80 f3Z (19.53 Hz) 78.57

Probability (%)
Probability (%)

64.29 64.29
60

40 35.71
28.57
20 f1Z (9.77 Hz) f2Z (12.50 Hz)

(b) f3Z (19.53 Hz)

(a) 0
1/1 2/2 3/3 4/4 5/5 1/1 2/2 3/3 4/4 5/5
Test/Span Test/Span

Figure 11. Dependence of the

Figure 11. probability
Dependence of of
theappearance
probabilityofofthe first threeoffrequencies
appearance identified
the first three in
frequencies identified in
both directions depending on the number of tests/span. (a) Direction Z, (b) Direction Y.
both directions depending on the number of tests/span. (a) Direction Z, (b) Direction Y.

To complete the Toidentification

complete the process, even ifprocess,
identification in an approximative way, the damping
even if in an approximative way, the damping
ratios in correspondence with the experimental
ratios in correspondence with the modes are reported
experimental modesin arethe following
reported Table
in the following Table 6.
6.
Table 6. Damping ratios identified in correspondence of the experimental modes.
Table 6. Damping ratios identified in correspondence of the experimental modes.
F = 9.77 Hz F = 12.50 Hz F = 18.18 Hz F = 19.53 Hz
F = 9.77 Hz Damping ratio
F = 12.50 Hz 2.00%
F = 18.18 1.00%
Hz F =2.80%
19.53 Hz 1.80%
Damping ratio 2.00% 1.00% 2.80% 1.80%

3. Finite Element Modeling

3. Finite Element Modeling
The bridge was modeled through a numerical finite element model implemented in
The bridge was modeled through a numerical finite element model implemented in
MIDAS Civil software. The whole model is composed of 548 nodes and 1129 elements
MIDAS Civil software. The whole model is composed of 548 nodes and 1129 elements
Buildings 2022, 12, x FOR PEER REVIEW (740 beams and 389 plates). The deck of the spans 1, 2, 4, and 5 is made of concrete,14for of 23
(740 beams and 389 plates). The deck of the spans 1, 2, 4, and 5 is made of concrete, for
3 ) with Young’s
which a class C20/25 was adopted (specific weight of 25 KN/m modulus
which a class C20/25 was adopted (specific weight of 25 KN/m 3) with Young’s modulus
assumed equal to 29.96 GPa. Instead, span n◦ 3 has a steel deck and so a steel class of S275
assumed equal to 29.96 GPa. Instead, span n° 33has a steel deck and so a steel class of S275
(specific weight 3of 76.98 KN/m ) with Young’s modulus of 210 GPa was assumed. All
(specific weight
basedof 76.98
on theKN/m ) with Young’s
information found modulus
in elastic of 210design
the original GPa was assumed. All
documents. Thecon-
density
constitutive models selected were and linear. The initial parameters wereof the mesh
selected
stitutive models selected
isbased
variable were
along elastic
the and linear.
structure. For The
plate initial
elements, parameters
on were
average, selected
there is an
on the information found in the original design documents. The density of the meshelement at each
14is m 2 but near the supports and over the steel central deck a refinement of 6.5 m2 and 3.5
variable along the structure. For plate elements, on average, there is an element at each
m14 m2 but near the
2, respectively, wassupports
carried and
out. over
In Figure 12 the
the steel 3D view
central deck of the implemented
a refinement 2 and is
of 6.5 mmodel
2
illustrated.
3.5 m , respectively, was carried out. In Figure 12 the 3D view of the implemented model is
illustrated.

Figure 12. The 3D view of the numerical and predictive model. The numbers indicate the spans on
Figure 12. The 3D view of the numerical and predictive model. The numbers indicate the spans on
which the experimental tests were carried out.
which the experimental tests were carried out.
The principal elements of the model are the following:
The principal elements of the model are the following:
• Concrete beams (Figure 13a,b): the box sections were modeled with 5 mono-dimensional
• Concrete beams (Figure
beam elements. 13a,b):
Each beam was the box sections
composed were modeled
of a 0.50-m-thick septumwith 5 mono-dimen-
and both extrados
sional beam elements. Each beam was composed of a 0.50-m-thick septum
and intrados slabs that were 0.20 m thick. The curvilinear development through andthe
both
extrados and intrados slabs that were 0.20 m thick. The curvilinear development
bridge spans length was realized with a total of about 200 sections of variable height.
• through
Concrete the bridgeelements:
transverse spans length wastwo
there are realized with
different a total
kinds of abouttransverse
of concrete 200 sections
ele- of
variable height.
ments (Figure 13c,d), both were modeled with mono-dimensional beam elements. The
• Concrete transverse elements: there are two different kinds of concrete transverse ele-
ments (Figure 13c,d), both were modeled with mono-dimensional beam elements.
The main transverse elements (Figure 13c) have different heights depending on the
deck while the thickness is constant and equal to 0.30 m. The secondary transverse
Buildings 2022, 12, 833 15 of 24

main transverse elements (Figure 13c) have different heights depending on the deck
while the thickness is constant and equal to 0.30 m. The secondary transverse elements
(Figure 13d) have constant sections of 0.75 m × 0.30 m.
• Steel deck (Figure 13e,f): the pairs of double T beams were modeled with 4 mono-
dimensional beam elements, having variable heights (in the end zones).
• Steel transverse elements (Figure 13g): there are 10 different types of steel transverse
elements. Each one was modeled as a mono-dimensional beam element with variable
sections.
• Cantilever steel beams (Figure 13h): the elements that support the steel deck on the right
side (reference in Figure 7) were modeled with 4 mono-dimensional beam elements,
with variable heights and rigidly connected to the pier.
•
Buildings 2022, 12, x FOR PEER REVIEW Deck slab: it was modeled with a 0.20-m-thick bidimensional plate element15for of 23
all
bridge spans.

Figure
Figure 13.
13. Representation
Representation of of
some
someimportant
importantmodel
modelelements: concrete
elements: longitudinal
concrete longitudinal beams: (a)
beams:
transversal section
(a) transversal and (b)
section andlongitudinal view.view.
(b) longitudinal Transversal beams:
Transversal (c) main
beams: (c) and (d)and
main secondary. Lon-
(d) secondary.
gitudinal beams
Longitudinal for the
beams for steel deck:
the steel (e) (e)
deck: transversal and
transversal (f)(f)
and longitudinal section;
longitudinal section;(g)
(g)steel
steeltransversal
transversal
elements; (h) cantilever steel beams; (i) deck slab; (l) model 3D view.
elements; (h) cantilever steel beams; (i) deck slab; (l) model 3D view.

The
Theboundary
boundaryconditions
conditionsfor forthe
themodel
modelare aredivided
dividedinto
intoexternal
externaland
andinternal
internal(end
(end
releases)
releases)ones
onesininFigure
Figure14a,b.
14a,b.The
Thefirst
firstare
arerestraints
restraintsthat
thatconnect
connectthe
thedeck
deckwith
withthe
thesub-
sub-
structures
structures(piles
(piles and
and lateral abutments).
abutments). The The end
end releases
releaseswere
wereapplied
appliedininthe
thecorresponding
correspond-
ing Gabble
Gabble saddles
saddles (R15–R
(R1 –R in Figure
in5 Figure 14b).
14b). In particular,
In particular, in each
in each node,
node, contained
contained in corre-
in correspon-
spondence with the transversal lines R –R removed the bending moments
dence with the transversal lines R1 –R5 removed the bending moments and longitudinal
1 5 and longitudi-
nal normal forces as indicated in Table 7. Both external boundaries conditions and beam
end releases were modeled consistently with what was reported in the original static
sketch. Moreover, to better represent the structural behavior of the bridge, a rigid plan
constraint was introduced through the function rigid link. All the nodes of the model were
Buildings 2022, 12, 833 16 of 24

normal forces as indicated in Table 7. Both external boundaries conditions and beam end
releases were modeled consistently with what was reported in the original static sketch.
Moreover, to better represent the structural behavior of the bridge, a rigid plan constraint
was introduced through the function rigid link. All the nodes of the model were connected
with a rigid link to a master node, positioned in the center of the whole bridge. To complete
the model implementation, the road pavement was taken into account using a uniformly
distributed load of 2 kN/m2 acting on the elements representative of the deck slab. The
Buildings 2022, 12, x FOR PEER REVIEW
numerical modal analysis was subsequently carried out and the results in terms16ofoffre- 23

quencies, participating mass, and modal shapes are reported in the following Table 8 and
Figure 15.

Figure 14.
Figure Highlightsofofthe
14.Highlights theboundary
boundary conditions:
conditions: (a)(a) external
external boundary
boundary conditions
conditions (hinges)
(hinges) andand
(b)
(b) beam end releases applied to simulate the Gerber saddles.
beam end releases applied to simulate the Gerber saddles.

Table 7.
Table Typologies of
7. Typologies of end
end release
release applied
applied to
to simulate
simulate the
the Gerber
Gerber saddles.
saddles.

Support
Support Fx
Fx Fy
Fy Fz
Fz Mx
Mx My
My Mz
Mz
R1
R1 00 11 11 11 00 00
R2
R2 11 11 11 11 00 00
R3 0 1 1 1 0 0
R3
R4 10 11 11 11 00 00
R5
R4 01 11 11 11 00 00
1: fixed; 0: free; Fx, Fy, Fz, Mx, My, Mz: forces and moments in the corresponding directions.
R5 0 1 1 1 0 0
1: fixed; 0: free; Fx, Fy, Fz, Mx, My, Mz: forces and moments in the corresponding directions.
The results of the numerical modal analysis related to an initial model defined using
nominal parameters (as, for example, the elastic modulus; in the following such initial
Table 8. Main modal frequencies and participating masses obtained by the nominal model.
model will be called the nominal model) are reported in Table 8. In particular, in grey the
Number of Frequency main modes
Part. Mass in Z-Part.
andMass
Y-directions, very closePart.
Part. Mass to the
Massidentified
Part. ones,
Mass are highlighted.
Part. Mass
Mode (Hz) Mode TX1(%)shows a bending
TY (%) deformed in correspondence
TZ (%) RX (%) with the RY central
(%) Gerber saddle
RZ (%)
1 9.73 (span 0.37
3) and, moreover,0.00 its frequency3.11(9.73 Hz) is 0.00
very close to the first identified
0.15 0.00mode.
2 10.79 The same
0.00 observation can be made for
0.01 0.00the third mode
0.01 (11.82 Hz) 0.00
that, moreover,0.00exhibits
a global shape. It could be reasonably associated to the second identified mode (12.50 Hz).
3 11.82 17.46 0.00 0.44 0.00 7.14 0.00
Instead, to find a mode close to the third identified mode (19.53 Hz), in both frequency
4 13.08 0.00 0.00 15.37 0.00 0.00 0.00
and shape, selection of the numerical mode 9 (16.17 Hz) is required. Mode 2 shows a local
5 13.77 9.95 0.00
and torsional deformed in the span 0.27 0.00
3. In this situation, 4.39
the corresponding 0.00
frequency was
6 14.21 0.00 0.01 0.00 0.01 0.00
not identified since the central measurements in that span (see position M3 in Figure 0.74 7)
7 14.22 0.01recorded correctly.
were not 0.15 0.00
Instead, mode 4 shows 0.30 0.01
modal displacements 0.050 and
in spans
8 16.05 6 where0.00measurements 0.00were not acquired.
16.74 Mode 50.00
is a global mode
0.00 of higher order
0.00 that
9 16.17 was not
0.79identified. In0.00
Figure 15 some PSDs that have
0.06 0.00frequencies0.33
associable to the
0.00corre-
10 16.93 sponding
0.00 numerical 93.05
modes are reported.
0.00 23.20 0.00 0.84
TY : translational Y-direction; TZ : translational Z-direction; RX : rotational (torsional) direction.
Table 8. Main modal frequencies and participating masses obtained by the nominal model.

Number of Frequency Part. Mass Part. Mass Part. Mass Part. Mass Part. Mass Part. Mass
Mode (Hz) TX (%) TY (%) TZ (%) RX (%) RY (%) RZ (%)
1 9.73 0.37 0.00 3.11 0.00 0.15 0.00
 Buildings 2022, 12, x FOR PEER REVIEW 17 of 23

8 16.05 0.00 0.00 16.74 0.00 0.00 0.00

9 16.17 0.79 0.00 0.06 0.00 0.33 0.00
Buildings 2022, 12, 833 17 of 24
10 16.93 0.00 93.05 0.00 23.20 0.00 0.84
TY: translational Y-direction; TZ: translational Z-direction; RX: rotational (torsional) direction.

Figure
Figure 15. Modal 15. Modal
shapes shapes corresponding
corresponding to the main
to the main modes modes
reported reported
in Table 8. in Table 8.

Manual
The results of theModel Calibration
numerical modal analysis related to an initial model defined using
nominal parameters Once (as,
thefor example,modal
numerical the elastic modulus;
frequencies in the following
are obtained such initial
by the nominal model, a com-
model will be nominal model)
parison with experimentally identified frequencies is opportune. The targetgrey
called the are reported in Table 8. In particular, in is to make as
the main modes closeinasZ-possible
and Y-directions, very
the difference close tonumerical
between the identified ones, are
frequencies andhighlighted.
identified ones. Such
Mode 1 shows a bending
objective deformed
was pursued in manual
with correspondence with the
model updates. It iscentral
importantGerber saddle
to focus the attention
(span 3) and,on moreover, its frequency
the numerical (9.73 Hz) isThe
modal frequencies. very close
three to the first
frequencies identified
that were takenmode.
into account
in the procedure
The same observation can be madeare reported
for the in Table
third 9 where
mode a first
(11.82 Hz) comparison
that, moreover,withexhibits
the numerical
a fre-
global shape.quencies
It couldcoming from theassociated
be reasonably nominal modelto theissecond
reported. The percentage
identified difference
mode (12.50 Hz).was eval-
uated
Instead, to find through
a mode closetheto following expression:
the third identified mode (19.53 Hz), in both frequency and
shape, selection of the numerical mode 9 (16.17 Hz) is required. 𝑓 − Mode 𝑓 2 shows a local and
torsional deformed in the span 3. In this𝐷𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒 situation, (%) =
the corresponding
𝑓
× frequency
100 was not (1)
identified since the central measurements in that span (see position M3 in Figure 7) were
According
not recorded correctly. Instead,to the
moderesults reported
4 shows modalin Table 9, the firstinnumerical
displacements mode,
spans 0 and in the Z-direc-
6 where
measurements were not acquired. Mode 5 is a global mode of higher order that wasand
tion, shows good agreement with the experimental one while the third one notespecially
the ninth have a greater difference. The average difference
identified. In Figure 15 some PSDs that have frequencies associable to the corresponding is 7.68%.
numerical modes are reported.
Table 9. Comparison between numerical (nominal model) and experimental frequencies.
Manual Model Calibration Numerical Model Experimental Results
Once the numerical modal frequencies
Mode Direction Frequency are(Hz)
obtained Frequency
by the nominal
(Hz) model, a com- (%)
Difference
parison with experimentally
Z identified frequencies
9.73 is opportune.9.77
The target is to make as
0.41
close as possible the difference
Z between numerical
11.82 frequencies and
12.50identified ones. Such
5.44
objective was pursued with
Z manual model updates. 16.17 It is important to focus the attention
19.53 on
17.20
the numerical modal frequencies. The three frequencies that were taken into account in the
procedure are reported in Table
Since the 9 where
comparison a first comparison
between the numericalwith theand
model numerical frequencies
experimental result frequen-
coming fromcies
the reveals
nominal model is reported. The percentage difference
relevant differences for the second and third modes (modes was evaluated
3 and 9 at 11.83
through the following
and 16.58 expression:
Hz, respectively), as shown in Table 9, the numerical model should be opti-
mized in order to minimize these differences.
Analyzing f exp − f num
Disuch frequencies,
f f erence (%) = it is possible×to100 observe that experimentally (1)identified
frequencies show higher values than fnumerical exp ones. The real structure appears stiffer
that the numerical one. In order to make the difference between the experimental and
numerical modal frequencies as small as possible, a stiffening of the numerical model is
Table 9. Comparison between numerical (nominal model) and experimental frequencies.
pursued by increasing the Elastic Modulus (E) of the concrete. The optimization criterion
Numerical Model Experimental Results
Mode Direction Frequency (Hz) Frequency (Hz) Difference (%)
Z 9.73 9.77 0.41
Z 11.82 12.50 5.44
Z 16.17 19.53 17.20
 Buildings 2022, 12, 833 18 of 24

According to the results reported in Table 9, the first numerical mode, in the Z-direction,
shows good agreement with the experimental one while the third one and especially the
ninth have a greater difference. The average difference is 7.68%.
OR PEER REVIEW
Since the comparison between the numerical model and experimental result frequen-
18 of 23
cies reveals relevant differences for the second and third modes (modes 3 and 9 at 11.83
and 16.58 Hz, respectively), as shown in Table 9, the numerical model should be optimized
in order to minimize these differences.
Analyzing such frequencies, it is possible to observe that experimentally identified
is driven using the following Objective Function (OF): minimizing the sum of the differ-
frequencies show higher values than numerical ones. The real structure appears stiffer
ences, in absolute value, of numerical
that the the threeone.
frequencies associated
In order to to the modes
make the difference selected
between in Table and
the experimental
9. numerical modal frequencies as small as possible, a stiffening of the numerical model is
pursued by increasing the Elastic Modulus (E) of the concrete. The optimization criterion is
𝑓 , Function
driven using the following Objective − 𝑓 , (OF): minimizing the sum of the differences,
𝑂𝐹 (𝐸) = min × 100to the modes selected in Table
in absolute value, of the three frequencies associated (2) 9.
𝑓,
!
3 f
i,exp − f i,num
It is correct to underline that in Equation
OF ( E) =(2)
minthe∑OF depends on the E parameter. In (2)
× 100
i =1
f i,exp
Figure 16 a parametric analysis of the frequencies altering the Elastic Modulus is reported.
This latter is increased Itwithin a range
is correct of 30%
to underline thatwith respect
in Equation (2) to
theits
OFnominal
depends on value
the E (29,961
parameter. In
MPa), as well visible in the abscissa. In the ordinate, according to Equation (1), the isper-
Figure 16 a parametric analysis of the frequencies altering the Elastic Modulus reported.
This latter is increased within a range of 30% with respect to its nominal value (29,961 MPa),
centage differences between numerical and experimental results are reported. The first
as well visible in the abscissa. In the ordinate, according to Equation (1), the percentage
(mode 1) and third differences
(mode 9) frequencies undergo
between numerical an increaseresults
and experimental and decrease, respectively.
are reported. The first (mode 1)
Instead, the second and
onethird
(mode 2) highlights a minimum point corresponding
(mode 9) frequencies undergo an increase and decrease, respectively. to 13%.Instead,
The the
red curve illustrates, instead,
second the sum
one (mode of sucha differences.
2) highlights minimum point The results indicate
corresponding to 13%.that thecurve
The red
optimum value of E is found increasing its nominal value up to 13%. Therefore, the final value
illustrates, instead, the sum of such differences. The results indicate that the optimum
of E is found increasing its nominal value up to 13%. Therefore, the final value inserted
value inserted in theinupdated model is equal to 33,856 MPa. Such value could be consid-
the updated model is equal to 33,856 MPa. Such value could be considered reasonable
ered reasonable since, in in
since, general,
general, E increases
E increases over
over timetime up
up to anto an asymptotic
asymptotic value. value.

Mode 1 Mode 3 Mode 9 Sum of the differences

Optimal value
Difference [%]

Elastic Modulus variation (E)[%]

Figure 16. ParametricFigure 16. Parametric

analysis increasing analysis increasing
the Elastic the Elastic
Modulus of Modulus of the numerical
the numerical frequencies
frequencies related to
related
to the modes 1, 3, andthe modes 1, shown
9. Results 3, and 9. in
Results shownof
the terms in the terms of percentage
percentage difference.
difference.
In Table 10 a comparison between the performance obtained by the initial (nominal)
In Table 10 a comparison
and updated between the performance
model is reported. Modes 3 andobtained by the model,
9, in the updated initialwere
(nominal)
considerably
and updated modelimproved
is reported. Modes
versus a slight3 detriment
and 9, inofthe
theupdated
first mode.model, were considerably
Consequently, the OF shifts from a
value of 23.05% to 10.28%.
improved versus a slight detriment of the first mode. Consequently, the OF shifts from a
value of 23.05% to 10.28%.
In order to complete the updated process, a comparison between experimental and
numerical shapes was also performed. The modal shapes (experimentally identified and
numerical) are reported in the Z-direction for modes 1 and 3 in the corresponding spans
1 and 3 in Figure 17.
 Buildings 2022, 12, 833 19 of 24

Table 10. Comparison between frequencies of the updated model and identified ones.

Identified E = 29,961 MPa (+0%) E = 33,856 MPa (+13%)

Frequencies Frequency Difference Frequency Difference
Mode/Direction (Hz) (Hz) (%) (Hz) (%)
Mode 1 Z 9.77 9.73 0.41% 9.91 −1.43%
Mode 3 Z 12.50 11.82 5.44% 12.54 −0.32%
Mode 9 Z 19.53 16.17 17.20% 17.18 12.03%

23.05% 10.28%

In order to complete the updated process, a comparison between experimental and

OR PEER REVIEW numerical shapes was also performed. The modal shapes (experimentally 19 of identified
23 and
numerical) are reported in the Z-direction for modes 1 and 3 in the corresponding spans
1 and 3 in Figure 17.

Span 1 – mode 1 Span 3 – mode 1

Modal coordinate

Num. Modal coordinate

Normalized

Exp. −
Normalized

−
− Num.
−
Exp.
− −

(a) (b)
Normalized Lenght Normalized Lenght

Span 1 – mode 3 Span 3 – mode 3

Modal coordinate

Modal coordinate

−
Normalized

Normalized

Num. − Num.
−
Exp. Exp.
− −

(c) (d)
Normalized Lenght Normalized Lenght

Figure 17. Comparison Figure

between17. experimental and numerical
Comparison between modal
experimental shapes: (a,b)
and numerical mode
modal 1 and
shapes: (a,b)(c,d)
mode 1 and (c,d)
mode 3 in corresponding spans
mode 3 in3corresponding
and 4. spans 3 and 4.

Notwithstanding the small number experimental modal coordinates (3 in span 1 and

Notwithstanding the small number experimental modal coordinates (3 in span 1 and
2 in span 3), a good agreement is clearly visible especially for mode 3 in span 3 where a
2 in span 3), a good agreement
counterphase isbehavior
clearly is
visible especially
recognizable in L3Zfor
andmode 3 inifspan
R3Z . Even it not3clearly
where a
visible, a slight
counterphase behavior is recognizable
deformation is shownin L
for and R
3Z span . Even
1,3Zmode 1. if
Initthis
notlast
clearly visible,
case, the a slight is missing
counterphase
deformation is shownespecially
for span 1, tomode
due 1. number
the low In this of
last case, the counterphase
measurements. The experimental is missing
modal shapes were
especially due to the low number of measurements. The experimental modal shapesthe
obtained through the AFDD procedure in Matlab. Finally, in Figure 18 weremodal shapes
obtained through therelated
AFDDtoprocedure
the modes 1,in3,Matlab.
and 9 of the updated
Finally, model were
in Figure inserted.
18 the modal It is right to highlight
shapes
that, after the model updating procedure, such modal shapes are not changed.
related to the modes 1, 3, and 9 of the updated model were inserted. It is right to highlight
that, after the model updating procedure, such modal shapes are not changed.
 counterphase behavior is recognizable in L3Z and R3Z. Even if it not clearly visible, a slight
deformation is shown for span 1, mode 1. In this last case, the counterphase is missing
especially due to the low number of measurements. The experimental modal shapes were
obtained through the AFDD procedure in Matlab. Finally, in Figure 18 the modal shapes
Buildingsrelated to the modes 1, 3, and 9 of the updated model were inserted. It is right to highlight
2022, 12, 833 20 of 24

that, after the model updating procedure, such modal shapes are not changed.

Figure 18. Modal shapes of18.

Figure theModal
updated
shapes model: (a,d,g)
of the updated 3D (a,d,g)
model: view;3D
(b,e,h) viewview
view; (b,e,h) from above;
from above; (c,f,i)
(c,f,i) lateral view.
lateral view.
4. Discussion

4. Discussion The tests on the bridge were performed in a one-day campaign and they highlighted
some difficulties. One is related to maintenance of the bridge in its operational condition,
The tests on theduring
bridge thewere
tests,performed
guaranteeinginthe a one-day
proper flowcampaign
of traffic. and they highlighted
The presence of the traffic if, from
some difficulties. One
theis related
side, madetothemaintenance
installation ofofthe
the bridge in its
experimental operational
layouts difficult;condition,
on the other side, it
allowed to record accelerations with an adequate amplitude level.
during the tests, guaranteeing the proper flow of traffic. The presence of the traffic if, from Moreover, the intensive
the side, made the installation of the experimental layouts difficult; on the other side,for
traffic did not make the realization of all useful configurations possible due to, it example,
the positioning of the sensors in the longitudinal center line of the roadway. Of course, the
tests were performed very quickly reducing the typical risks related to the execution of
the dynamic tests. A first general suggestion for the execution of vibrational experimental
tests on bridges is the following: the use of wired instrumentation raised issues due to the
deployment of the communication cables and so, especially for the experimental activity
on bridges or viaducts, a wireless sensor nodes network is advisable. However, in the last
case, an opportune choice of the Micro Electro-Mechanical System (MEMS) sensor, with
sufficient sensitivity, is needed. The results from the use of recursive/multiple tests suggest
that, especially for bridges showing a very long longitudinal configuration, they could
provide useful indications. Indeed, a statistical approach helps with removal of possible
uncertainness related to few measurements available and the complexity of the structure.
It was also found that the peaks in the acquired time histories were due to the passage
of heavy vehicles on the bridge. This observation suggests that the number of times in
which the accelerations overcome a well-defined threshold is equivalent to the same number
of heavy vehicles transiting. In old bridges, as in the case study analyzed in this paper,
it could be important to know statistical information related to the passage of particular
typologies of vehicles and monitor the evolution over time.
Therefore, the installation of a vibration-based Structural Health Monitoring (SHM)
system would allow the acquisition of data from which it would be possible to extrapolate
information useful for evaluating the health of the structure and extend its cycle life. Moreover,
such system could also be endowed an automatic or semi-automatic procedure able to send
messages to human operators in the presence of malfunctions. One of the aims of the
SHM systems is to identify the structural damage due to both degradation of the material
or after a heavy event such as an earthquake. In such way, some advantages could be
highlighted: (1) minimization of the human errors and safer conditions for technical operations,
(2) regularity and quickness in obtaining the information, (3) higher reliability related to the
actual state of the infrastructure and therefore higher confidence in decision making.
 Buildings 2022, 12, 833 21 of 24

While SHM systems for industrial infrastructures have now been standardized and
marketed, in the case of civil ones, an identical approach is not always applicable. In
these latter situations, a SHM system needs to be accurately designed (in some sense “cus-
tomized”) based on the targets to be pursued and the type of structures or infrastructure.
The instrumentation and specific operating management procedures require interdisci-
plinary knowledge (e.g., civil, electronic, and informatic). Therefore, in order to obtain an
optimization of the whole process, specialized technical staff, with different skills, are nec-
essary. The adoption of a permanent monitoring strategy, in which the hardware/software
system is designed to remain operational for long periods to cover the entire service life
of a facility, fully achieves the purposes of the Structural Health Monitoring system and
full-scale integration with surveillance activities.

Preliminary Definition of a Continuous Monitoring System: Example for the Case Study
In the design of monitoring systems concerning bridges, particular attention must be
12, x FOR PEER REVIEW paid to the problems of durability, robustness and maintainability of sensors,
21 of 23 electronic
data acquisition, and transmission equipment [29]. In particular, the following aspects
must be carefully considered:
• System architecture.
• Reliability and
• insensitivity
Redundancyto electromagnetic
and flexibility of the disturbances
sensor network. of equipment and data
transmission•lines.Accuracy of the measurement system and reliability of data processing.
• Specific rules• of acquisition
Power supply (sample time,
(operative alert
even thresholds).
in the case of critical events such as an earthquake).
• •
Size of the databases containing measures and data management
Reliability and insensitivity to electromagnetic algorithms
disturbances (big and data
of equipment
data). transmission lines.
• Specific rules of acquisition (sample time, alert thresholds).
Based on these assumptions, a basic continuous monitoring system was developed
• Size of the databases containing measures and data management algorithms (big data).
for the case study. A possible planimetric arrangement of the accelerometers is shown in
Based on these assumptions, a basic continuous monitoring system was developed
Figure 19. The configuration is coherent with the layouts implemented in the dynamic
for the case study. A possible planimetric arrangement of the accelerometers is shown in
experimental tests described
Figure 19. Theabove. Substantially,
configuration forwith
is coherent eachthespan, siximplemented
layouts sensor positions
in the dynamic
(three in each side) were designed:
experimental two in the
tests described centerline,
above. and the
Substantially, forother
each two
span,couple, respec-
six sensor positions (three
tively, in one third andside)
in each two were
thirddesigned:
of the whole
two in span length. and
the centerline, Moreover,
the othereach position,
two couple, respectively,
should be endowed tridimensional sensors to have a sufficient data and information.
in one third and two third of the whole span length. Moreover, each position, should be
endowed tridimensional sensors to have a sufficient data and information.

Figure 19. Configuration

Figure of
19.aConfiguration
possible SHM ofsystem designed
a possible for “Ponte
SHM system della
designed forMagliana”.
“Ponte della Magliana”.

Two different
Two different hypotheses hypotheses of accelerometers
of accelerometers could be adopted:
could be adopted:
• Piezoelectric uniaxial accelerometers.
• Piezoelectric uniaxial accelerometers.
• MEMS accelerometer.
• MEMS accelerometer.
The accelerometers could be installed inside concrete shafts and connected to a data
The accelerometers could
acquisition be installed
system inside
by coaxial cablesconcrete shafts along
that are spread and connected to a data
the development of the bridge
acquisition system by coaxial cables
within corrugated pipes.that are spread along the development of the bridge
within corrugated pipes. In the following Tables 11 and 12 some general indications are reported for the instal-
lationTables
In the following costs of
11a and
permanent
12 somemonitoring system usingare
general indications both typologies
reported for of
theaccelerometers
in-
selected above.
stallation costs of a permanent monitoring system using both typologies of accelerometers
selected above. The installation of a monitoring system could be considered a start point for the
development of a “Digital Twin” model useful to realize a connection between the real and
digital infrastructure.
Table 11. General costs for a SHM system on Magliana bridge using Piezoelectric accelerometers.

Element Unit Price Unit Total

Piezoelectric uniaxial accelerometers EUR 1300.00 84 EUR 109,200.00
Connecting coaxial cable EUR 10.00 446 EUR 4460.00
Buildings 2022, 12, 833 22 of 24

Table 11. General costs for a SHM system on Magliana bridge using Piezoelectric accelerometers.

Element Unit Price Unit Total

Piezoelectric uniaxial accelerometers EUR 1300.00 84 EUR 109,200.00

Connecting coaxial cable EUR 10.00 446 EUR 4460.00

Additional tools EUR 9239.14

Total EUR 122,899.14

Table 12. General costs for a SHM system on Magliana bridge using MEMS accelerometers.

Element Unit Price Unit Total

MEMS triaxial accelerometers EUR 1162.83 42 EUR 48,838.86

Acquisition and processing data system EUR 3827.08 1 EUR 3827.08

Additional tools EUR 9239.14

Total EUR 61,905.08

5. Conclusions
The paper contributes to defining effective procedures for fast and low cost experi-
mental dynamic testing discussing the interesting case of a multi-span arch bridge. The
peculiarities in the execution of the tests for long-span bridges using a low number of
sensors were highlighted. In order to analyze the global dynamic behavior, multiple ex-
perimental tests were carried out. The idea was suggested also thanks to the regularity of
the infrastructure that showed a repetitive configuration (arched-shapes spans linked by
Gerber saddles). Throughout the performed activities, several observations can be made:
• Multiple and recursive tests could constitute a possibility for obtaining reliable infor-
mation using dynamic measurements, especially in the case in which the infrastructure
shows a geometric regularity and a linear development.
• A statistical approach to carry out a modal identification is always advisable and useful
to reduce the uncertainties. In the example illustrated, it was possible to recognize
the frequencies associated to global modes. Some other critical issues, not completely
eliminable by a statistical analysis, could be solved coupling the results with the ones
obtained using a reliable numerical model.
• A numerical model updated using the information from the experimental test could
also be used to design an appropriate SHM system. Moreover, the vibrational informa-
tion, provided by such system, could be processed also for achieving other information
useful for optimal management of the infrastructure (e.g., to monitor the intensity and
typology of the traffic flow).
Certainly, limitations should be also highlighted:
• The correct identification of the global dynamical behavior needs a large number of
measurements. They must be accurately processed and the application expeditious
procedures could be extensive and slow to manage.
• Some defects are not directly detectable by vibrational global information since they
are due to local anomalies or another nature (e.g., vegetation, efflorescence, corrosion).
In these situations, it is advisable to undertake a data fusion of information coming
from different technologies (for example, image processing).
• Another important issue will be related to developing a complete automatized proce-
dure using a robotized system.
Buildings 2022, 12, 833 23 of 24

Author Contributions: V.G.: conceptualization, supervision, funding acquisition; F.P.: conceptualiza-

tion, field measurements and identification, writing and review editing; G.P.: structural modelling,
numerical investigation and original draft preparation. All authors have read and agreed to the
published version of the manuscript.
Funding: Part of the research leading to these results has received funding from the Italian Govern-
ment under Cipe resolution n.135 (21 December 2012) under the project INCIPICT—INnovating City
Planning through Information and Communication Technologies and research project DESDEMONA—
DEtection of Steel Defects by Enhanced MONitoring and Automated procedure for self-inspection and
maintenance (grant agreement number RFCS-2018_800687) supported by EU Call RFCS-2017.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Acknowledgments: The authors wish to thank Eng. Umberto Di Sabatino for his assistance during
the experimental tests.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Galvin, P.; Romero, A.; Moliner, E.; De Roeck, G.; Martinez-Rodrigo, M.D. On the characterisation of railway bridges through
experimental testing. Eng. Struct. 2021, 226, 111261. [CrossRef]
2. Foti, D.; Giannoccaro, N.I.; Vacca, V.; Lerna, M. Structural Operativity Evaluation of Strategic Buildings through Finite Element
(FE) Models Validated by Operational Modal Analysis (OMA). Sensors 2020, 20, 3252. [CrossRef] [PubMed]
3. Brinker, R.; Zhang, L.; Andersen, P. Modal identification of out-only systems using frequency domain decomposition. Smart
Mater. Struct. 2001, 10, 441–445. [CrossRef]
4. Peeters, B.; De Roeck, G. Reference-based stochastic subspace identification for output-only modal analysis. Mech. Syst. Signal
Process. 1999, 13, 855–878. [CrossRef]
5. Valvona, F.; Toti, J.; Gattulli, V.; Potenza, F. Effective seismic strengthening and monitoring of a masonry vault by using Glass
Fiber Reinforced Cementitious Matrix with embedded Fiber Bragg Grating sensors. Compos. B Eng. 2017, 113, 355–370. [CrossRef]
6. Helmerich, R.; Niederleithinger, E.; Trela, C.; Bien, J.; Kaminski, T.; Bernardini, G. Multi-tool inspection and numerical analysis of
an old masonry arch bridge. Struct. Infrastruct. Eng. 2010, 8, 27–39. [CrossRef]
7. Beben, D.; Ukleja, J.; Maleska, T.; Anigacz, W. Study on the Restoration of a Masonry Arch Viaduct: Numerical Analysis and Lab
Tests. Materials 2020, 13, 1846. [CrossRef]
8. Gattulli, V.; Cunha, A.; Caetano, E.; Potenza, F.; Arena, A. Dynamical models of a suspension bridge driven by vibration data.
Smart Struct. Syst. 2021, 27, 139–156.
9. Dallard, P.; Fitzpatrick, A.J.; Flint, A.; Le Bourva, S.; Low, A.; Ridsdill Smith, R.M.; Willford, M. The London Millennium
Footbridge. Struct. Eng. 2001, 79, 17–33.
10. Nie, Z.; Guo, E.; Li, J.; Hao, H.; Ma, H.; Jiang, H. Bridge condition monitoring using fixed moving principal component analysis.
Struct. Control Health Monit. 2020, 27, e2535. [CrossRef]
11. Alijani, A.; Abadi, M.K.; Ghyasvand, S. Numerical and experimental analysis of intact and cracked railway bridges. Structures
2022, 38, 1018–1030. [CrossRef]
12. Silva, R.; Costa, C.; Arede, A. Numerical methodologies for the analysis of stone arch bridge with damage under railway loading.
Structures 2022, 39, 573–592. [CrossRef]
13. Foti, D.; Gattulli, V.; Potenza, F. Output-Only Identification and Model Updating by Dynamic Testing in Unfavorable Conditions
of a Seismically Damaged Building. Comput.-Aided Civ. Inf. 2014, 29, 659–675. [CrossRef]
14. Gattulli, V.; Lofrano, E.; Paolone, A.; Potenza, F. Measured properties of structural damping in railway bridges. J. Civ. Struct.
Health Monit. 2019, 9, 639–653. [CrossRef]
15. Chen, G.-W.; Omenzetter, P.; Beskhyroun, S. Operational modal analysis of an eleven-span concrete bridge subjected to weak
ambient excitations. Eng. Struct. 2017, 151, 839–860. [CrossRef]
16. Ni, Y.C.; Alamdari, M.M.; Ye, X.W.; Zhang, F.L. Fast operational modal analysis of a single-tower cable-stayed bridge by a
Bayesian method. Measurement 2021, 174, 109048. [CrossRef]
17. Xu, Y.; Brownjohn, J.M.W.; Hester, D. Enhanced sparse component analysis for operational modal identification of real-life bordge
structures. Mech. Syst. Signal Process. 2019, 116, 585–605. [CrossRef]
18. Benedettini, F.; Gentile, C. Operational modal testing and FE model tuning of a cable-stayed bridge. Eng. Struct. 2011, 33,
2063–2073. [CrossRef]
19. Benedettini, F.; Dilena, M.; Morassi, A. Vibration analysis and structural identification of a curved multi-span viaduct. Mech. Syst.
Signal Process. 2015, 54–55, 84–107. [CrossRef]
Buildings 2022, 12, 833 24 of 24

20. Caglayan, B.O.; Ozakgul, K.; Tezer, O. Assessment of a concrete arch bridge using static and dynamic load tests. Struct. Eng.
Mech. 2012, 41, 83–94. [CrossRef]
21. Conde, B.; Ramos, L.F.; Oliveira, D.V.; Riveiro, B.; Solla, M. Structrural assessment of masonry arch bridge by combination of
non-destructive testing techniques and three-dimensional numerical modelling: Application to Villanova bridge. Eng. Struct.
2017, 148, 621–638. [CrossRef]
22. Gentile, C. Modal and structural identification of a R.C. arch bridge. Struct. Eng. Mech. 2006, 22, 53–70. [CrossRef]
23. Gara, F.; Nicoletti, V.; Carbonari, S.; Ragni, L.; Dall’Asta, A. Dynamic monitoring of bridges during static load tests: Influence of
the dynamics of trucks on the modal parameters of the bridge. J. Civil Struct. Health Monit. 2020, 10, 197–217. [CrossRef]
24. Ataei, S.; Miri, A. Investigating dynamic amplification factor of railway masonry arch bridge through dynamic load tests. Constr.
Build. Mater. 2018, 183, 693–705. [CrossRef]
25. Gare Comune Roma. Available online: https://gare.comune.roma.it/gare/id2859-dettaglio (accessed on 11 January 2021).
26. Bendat, J.S.; Piersol, A.G. Engineering Applications of Correlation and Spectral Analysis, 2nd ed.; Wiley Interscience: Amsterdam, The
Netherlands, 1993.
27. Golub, G.H.; Van Loan, C.F. Matrix Computation, 3rd ed.; John Hopkins University Pres: Baltimore, MD, USA, 1996.
28. Cheynet, E.; Jakobsen, J.B.; Snæbjörnsson, J. Damping estimation of large wind-sensitive structures. Procedia Eng. 2017, 199,
2047–2053. [CrossRef]
29. Ministero delle Infrastrutture e dei Trasporti. Consiglio Superiore dei Lavori Pubblici Linee Guida per la Classificazione e Gestione del
Rischio, la Valutazione Della Sicurezza ed il Monitoraggio dei Ponti Esistenti; Ministero delle Infrastrutture e dei Trasporti: Rome, Italy,
2020; pp. 74–78.