Professional Documents
Culture Documents
DOC-20240528-WA0017.
DOC-20240528-WA0017.
DOC-20240528-WA0017.
To cite this article: G. Standoli, E. Giordano, G. Milani & F. Clementi (2020): Model Updating of
Historical Belfries Based on Oma Identification Techniques, International Journal of Architectural
Heritage, DOI: 10.1080/15583058.2020.1723735
Article views: 50
a
Department of Civil and Building Engineering, and Architecture, Polytechnic University of Marche, Ancona, Italy; bDepartment of
Architecture, Built Environment and Construction Engineering, Polytechnic of Milan, Milan, Italy
CONTACT F. Clementi francesco.clementi@univpm.it Department of Civil and Building Engineering, and Architecture, Polytechnic University of
Marche, Ancona, Italy
© 2020 Taylor & Francis
2 G. STANDOLI ET AL.
Preciado 2015) and on Discrete Element Model (DEM) influence of environmental actions such as temperature
(Coïsson et al. 2016; Ferrante, Clementi, and Milani and humidity, thus it will be possible to make the data
2019), to identify their vulnerability, and to design proper comparable with subsequent monitoring, assessing the
interventions. The results of these enquiries seem to truly state of preservation of the structure.
represent buildings’ behavior under seismic loads, even Meanwhile, the second type is executed in a period
though it is necessary that models also replicate the real that may last years, to analyze the variation of the
dynamic characteristics to grant a response that is related behavior during time and therefore to evaluate the
to the facts. For this reason, knowing the dynamic per- health of a dynamical system by validating the modal
formance may be seen as the starting point of the study parameters. During this test it would be appropriate to
(Gentile, Guidobaldi, and Saisi 2016; Quagliarini, also monitor wind, humidity and temperature, to inves-
Maracchini, and Clementi 2017) tigate how particular circumstances or ambient factors
So, being the matching between analytical and (Kita, Cavalagli, and Ubertini 2019) such as freezing
experimental models of analysis deeply correlated to affect the dynamic response.
a correct definition of the properties of the analyzed In this paper, the short-time AVT methodology is
buildings (geometry, mechanical properties, cracking applied in order to carry out the dynamic identification
conditions, etc.), it is necessary to establish these often- of four masonry bell towers. To highlight dynamic para-
uncertain parameters with a good grade of confidence. meters the acquired time histories were studied both in
In this sense, non-destructive identification techniques, time and in frequency domains, with the aim to compare
based on vibrational monitoring of structures, appear the two methodologies. The obtained parameters were
to be a useful tool to identify the dynamic behavior of then used to update the NMs of each tower. As other
structures. authors did (Bayraktar et al. 2009; Foti et al. 2012;
Two experimental methodologies are currently avail- Gentile and Saisi 2007), the process of model updating
able: (i) the Experimental Modal Analysis (EMA), and (ii) was conducted focusing on the modification of some
the Operation Modal Analysis (OMA). The first one uncertain materials’ parameters, in particular Young’s
consists in applying to the structure a known force and Modulus, in order to have a match of the first natural
monitoring its response; it is mostly used for laboratory frequencies and mode shapes between FEMs and OMA.
tests (Lenci, Consolini, and Clementi 2015), whereas it is The towers studied are situated in the northern part of
hardily applied in existing buildings (Laura, Francesco, Italy in the Emilia-Romagna region, precisely in Ferrara
and Antonio 2019), especially on a CH building, due to province (Figure 1), and they are:
the possibility to generate cracks and considering the
dimensions and costs of the input machine (Bartoli et al. ● San Giorgio Cathedral Belfry in Ferrara;
2014, 2020; Facchini and Betti 2016; Korumaz et al. 2017). ● San Benedetto Church Belfry in Ferrara;
On the other hand, the Operation Modal Analysis (OMA) ● Matildea Tower of the Santa Maria church in
consists on the Ambient Vibration Testing (AVT), which Bondeno;
involves the monitoring of the vibrations made on the ● Pomposa Abbey Belfry in Codigoro.
constructions by ambient sources; it is the most used
method for CH, since it grants the possibility to study
a building in its full scale (Gentile and Saisi 2007), analyz- 2. Geometrical and historical description of the
ing its dynamical behavior during its normal conditions case studies
of usage, without applying an artificial excitation or inter-
2.1. San Giorgio cathedral belfry
fering with its operativity (Pierdicca et al. 2019; Stefano,
Matta, and Clemente 2016; Ubertini, Comanducci, and The first presented bell-tower rises in front of the
Cavalagli 2016). Trento-Trieste square in the old town of Ferrara, on
The AVT can be carried out with different purposes, the south side of the cathedral. The cathedral dates
referring to two principal methods: (i) the shorth-time, back to the early XII century, while the actual belfry’s
and (ii) the long-time monitoring. The first one is building process started only in 1412 under the gov-
carried out in a few days and it is executed to identify ernment of Nicolò III d’Este, after that seismic events
the behavior of the structure in that particular time; of 1495 and 1570 damaged the old two little towers
information can be used to upload the Numerical that stood on the sides of the presbytery. The struc-
Model (NM) in order to have a correct representation ture was erected between 1412 and 1844 but, due to
of a building’s behavior to properly evaluate the real economic matters, the construction was never com-
seismic risk and the potential interventions. To make pleted, so it still lacks the initially designed cuspid on
the data truly reliable it is essential to eliminate the the roof. During the years the structure experienced
INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 3
different changes, considering that it was born as direction; the inclination is mainly caused by the
isolate tower but in 1703 at the behest of Papa sandy soil where the city rises up.
Clemente XI a choir was built beside it and a series
of horizontal passages were realized on the vertical
2.2. San Benedetto belfry
walls. However, the bell-tower returned to be isolated
in 1944, after a bombing during the II Word-war that The isolated belfry of San Benedetto Church arises in
struck down the choir. The tower has a square plan of the homonymous Square in Ferrara and it is located on
11.70 m sides and a maximum height of 50.78 m. By the North-East side of the Church, whose construction
observing the structure, we can notice four overlap- started in 1496 thanks to Ercole I d’Este and Andria
ping modules (nuts) whose sides length, constant Bishop after that Benedictine monks abandoned
from the bottom to the top, is approximately 11.7 m. Pomposa Abbey because of a malaria epidemy. The
The modules are separated by majestic trabeations, belfry was built starting from 1621, based on the design
supported by enormous corner columns. The inter- proposed by Giovanni Battista Aleotti, and was con-
mediate floors of the two first modules are in wooden, cluded in 1646. During World War II, the church was
while the others are made by masonry cross vaults, destroyed but the tower remained unscathed. During
except for the 6th, which has a masonry barrel vault. the years the bell tower was damaged by a series of
The cover floor is a not accessible wooden pavilion. storms, the most serious one occurred in 1842 provok-
The walls are 1.2 m thick, constituted by bricks ing the collapse of the upper part. The structure was
clamped following the state of the art, with aligned also damaged by the seismic events of 2012 that lead to
horizontal courses made with good quality mortar. the opening of numerous vertical cracks in both North
The bricks are covered with rose and white limestone. and South walls.
The thickness remains constant for all the vertical The tower is 52. 06m tall and presents a squared
development of the belfry. In all the modules, on the cross-section plan (7.33 x 7.33 m), which remains con-
southern side of the tower we can notice arc openings stant till the last floor (where the second bell cell is
(two for each floor). The arcs are supported by circu- located), where the section decreases to the size of
lar columns with diameters varying between 0.75 and 5.7 × 5.7 m. Even though the outer perimeter remains
0.9 m. The same architectural order is present on the constant up to the second floor, the walls taper on the
other sides but without any openings. Only the last inside from 1.40 m to 0.88 m, while the last floor they
module has a different style, with wider arc openings consist of 0.45 m. There are only two masonry cross
located on every side of the bell cell. On the west side vault slabs inside the tower, respectively to 31.37 m and
of the tower, there are remains of the old choir of the 39.78 m, connected by a steel stair (Figure 3).
church (Figure 2). As many of the towers in Ferrara, The walls are made of solid and regularly disposed
this one is not in perfect balance since it is inclined of bricks, and with an effective joint offset. Like the one
2° in East-West direction and 0.5° in North-South previously described, this tower also has an evident
4 G. STANDOLI ET AL.
inclination (Milani, Shehu, and Valente 2017), due to (with a height of 12 m) and it was not meant to be
the development of foundation’ settlement right after a belfry, but it had the aim both to sight and to defend.
the start of the construction works. The building is Two other floors were added later thus the tower
tilted by 0.5° along North-South direction and 3.07° reached a height of 29.96 m becoming a belfry. This
along East-West direction, which leads to development is also testified by the last-gothic style of
a displacement of the top of the tower of 0.49 m in the bell cell, typical of the latest part of the XIV century.
the North-South direction and 2.82 m in the East-West The structure of the tower has a squared cross-section
one. The inclination increased after the 2012 Emilia that is constant up to the last floor, with sides of 7.2 m
earthquakes, as reported by the study of (Pellegrinelli, long and a maximum height of 29.96 m. On the north
Furini, and Russo 2014) that identified an increase of and east sides, the tower is linked to the church by thin
the upper displacement of 18 mm. After the 2012 earth- walls. For what concerns the floors constructive typol-
quake a series of steel chains were installed at different ogy, there are cross masonry floors on the first two
levels to secure the structure and the foundation was levels, made with solid bricks, while in the third and
reinforced using micro-plies to prevent a further the fourth levels the floors are made of wood with
increase of the inclination. orthogonal warping. The cover floor is formed by visible
wooden trusses. The walls of the tower are made of solid
bricks, with a double case structure that accommodates
2.3. Matildea tower
the stairs in its middle. The external thickness is around
Matildea belfry arises on the side of Maria Vergine 1.05 m for the first three floors, while it decreases to
Church, which is the main religious building in the 0.9 m in the remaining three. For the inner part of
city of Bondeno. The church was built in 1114 thanks masonry, the thickness is constant and equal to 0.45 m.
to a donation made by Matilde of Canossa. The tower’s The bell cell, that is free of bells, is characterized, on
building process started around the XII century and was every side, by ogival arcs that encase three-light windows
concluded at the beginning of the XV century. At the with pointed arches. The ogival arches are supported by
beginning the tower was composed by only two floors little columns, surmounted by oculi (Figure 4).
INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 5
2.4. Pomposa Abbey belfry pyramid base composed of marmoreal elements. The
external walls are made of red and yellow solid bricks,
This tower is part of the complex of the Benedictine
with some fragments of marmoreal parts. Its thickness
monastery of Pomposa, located in the municipality of
varies from 1.34 m to 0.59 m. Openings of different shapes
Codigoro, and it is a typical example of Romanic archi-
are presents on the facades, their size and the number
tecture. The abbey was built on the remains of a little
increasing up to the top in order to decrease the weight
church of which the construction year is unknown, even
of the structure. The floors are made of wood with ortho-
though the first traces of its presence date back to IX
gonal warping at all levels, except for the bell cell floor
century, while the bell-tower was built-in only 1063 by
where a concrete ceiling is present. The levels are con-
architect Deusdedit. The belfry did not undergo through
nected through wooden stairs, and the tower is connected
many modifications through the years: the first restora-
at its base to the monastery by thin walls (Figure 5).
tion work dates in 1879, when it was necessary to recon-
struct the roof cone because of its failing caused by
a lightning bolt; in this occasion the walls were also
3. Ambient vibration testing
reinforced through the application of steel chains, posed
at floors’ levels. During the last years a foundation settle- This type of structures, along with all the historical
ment caused the inclination of the building due to the masonry buildings, is usually studied by applying linear
weight of the structure (around 1800 tons). and nonlinear analysis on FEMs, since Numerical
The belfry is 4 9.90m tall, with a squared cross-section, Models (NMs) give the chance to operate significant
whose sizes at the base are 7.70 × 7.70 m degressive with simplifications on historical structures due to the impos-
the height. The structure is composed by nine modules, sibility to clearly identify the spandrels and piers for
surmounted by a conical shape cover of 11.69 m in height their irregular geometry (Quagliarini, Maracchini, and
and 6.35 m in diameter. The building lands on a truncated Clementi 2017). In most cases, FEMs seemed to be able
6 G. STANDOLI ET AL.
to reproduce the real behavior but to be sure on the anti-aliasing filter related to a multi-channel data acqui-
accuracy of the results all the parameters as the proper- sition system CompacDAQ-9132.
ties of materials and each connection between the ele- The configuration was set in the most favorable way
ments should be known. During the last few years, the to detect the translational and torsional components of
most common technique used to overcome these issues displacements of the towers (Ubertini et al. 2017). With
is the AVT, which consists on the monitoring of vibra- this aim, the sensors were positioned to the corners or
tions of the buildings under ambiental noises as wind, close to them, depending on the accessibility of the
traffic etc., in order to identify the real dynamic beha- areas. To acquire the largest number of measurements
vior. The monitoring is performed by using acceler- with only a few accelerometers, different acquisitions
ometers placed on the most relevant positions of the were carried out by keeping some sensors fixed in the
structure that are probably most involved by the motion. highest reachable position for the towers and modifying
After recording and filtering the data, the results are the positions of other sensors at each acquisition. The
processed and then used to produce an Experimental accelerometers were fixed with bi-component resin to
Model (EM). The EM response is compared with the make sure they were properly connected to the struc-
preliminary response of the NM, which is then refined, ture, and to prevent damages on the building. The
by focusing on the unclear parameters, in order to fit the different sensors layout is shown in Figure 6.
experimental results. In this section each phase is shown The registration time for vibrations varied from
starting by the monitoring campaign, passing through 20 min to over 1 h, in order to respect Rodriguez
the analysis of the data and to finish with the construc- condition (Rodrigues 2004) which imposes an acquisi-
tion and calibration of the NMs. tion’s duration of at least 2000 times the expected
period for the structure, so capturing was long enough
to eliminate the influence of possible non-stochastic
3.1. Acceleration sensors layout and data excitations that may occur during the tests. Raw data
acquisition settings were originally acquired with a sample rate of 1000 Hz.
The sensor’s network used in the survey campaign
was composed of 8 uniaxial accelerometers, with
3.2. AVT data processing
a sensitivity of 1V/g and a measurement range of 8g,
connected to two 4-channels modules NI9132 AC/DC The data were elaborated to obtain the dynamic char-
with a 24-bit resolution, 102dB dynamic range and acteristics. The first operation was to clean the data by
INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 7
all the noise applying a low pass filter. Then, in order to and signal noise, the occurrence of aliasing and leakage
decrease the computational time, raw data were re- in the signal post-processing, identification errors, or
sampled through a decimation, before the real signal asymmetries linked to gyroscopic effects or non-linear
processing, obtaining the final range of 0–6.25 Hz, con- behavior, it is possible to incur in complex modes
sistent with the frequencies range of the analyzed (Bajrić and Høgsberg 2018). Consequently, the criteria
structures. adopted for the selection of the modal frequencies was
Different algorithms are suitable to process the that the chosen ones should have been those with the
data; in this paper the identification of the frequencies lowest level of complexity, which can be measured
of the four belfries was conducted through two differ- through the calculation of the so called Mode
ent methods, the Enhanced Frequency Domain Complexity Factor (MCF) (Brincker, Zhang, and
Decomposition (EFDD) and the Stochastic Subspace Andersen 2000; Ewins 1986):
Identification (SSI), in order to compare the results
2
(Torres et al. 2017). Sxx Syy þ 4S2 xy
In the application of both the methods, we realized MCFr ¼ 1 2 (1)
Sxx þ Syy
that the ideal condition in the definition of a mode is
that the eigenvector that describes the mode shape is
where:
composed of all real values: satisfying this condition
means that for every vibration cycle, being the phase
● r subscript
equal to 0° or 180°, the maximum deflection of all the indicates
T the considered mode;
parts of the structure is reached at the same time and so ● Sxx ¼ Re ψ r Re ψ r ;
T
it is possible to adopt a proportional damping model. ● Syy ¼ Im ψ r Im ψ r ;
T
However, because of the effect of measurement errors ● Sxy ¼ Re ψ r Im ψ r .
8 G. STANDOLI ET AL.
3.2.1. Enhanced frequency domain decomposition ● ðωÞ is the m r Frequency Response Function
H
The EFDD is an improved version of the classical (FRF) matrix, the overbar indicates its complex
Frequency Domain Decomposition (FDD) technique, conjugate and while the superscript T is the
which uses a single line from the Fast Fourier transpose.
Transform (FFT) for the estimation of the natural fre-
quency and does not calculate the modal damping. The
The Gyy matrix is composed by the vectors of the
EFDD method consists in identifying a Single Degree of
modal shapes associated to the modes concurring at
Freedom (SDOF) Power Spectral Density (PSD) func-
a particular frequency. Through a process known as
tion in the proximity of a peak of resonance, high-
Singular Value Decomposition (SVD), this form of
lighted through the Peak Picking method (like in
the matrix is decomposed into a set of eigenvalues
FDD), and then transporting it back to the time
and their corresponding eigenvectors, which are an
domain with the Inverse Discrete Fourier Transform.
approximation of the mode shapes. This decomposition
In EFDD method the measured responses are linked to
process leads to the identification of the SDOF models
the unknown input through the relationship expressed by:
of the problem (Gade et al. 2005). Then the natural
ðωÞ½Gxx ðωÞ½H
Gyy ðωÞ ¼ ½H ðωÞT (2) frequency is calculated on the base of the number of
zero-crossing as dependent from time, while the modal
where: damping is evaluated from the logarithmic decrement
of the normalized correlation function associated to the
● Gxx is the r r PSD matrix of the r inputs; SDOF system. Considering also modal damping, EFDD
● Gyy is the m m PSD matrix of the m responses; allows a more accurate estimation of both natural
INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 9
frequencies and mode shapes, in comparison with FDD most likely generated by its high inclination, in this case
(Benedettini, Dilena, and Morassi 2015). the distance among the first two frequencies is 72% while
In Figure 7 the singular values of the PSD matrix of the third is also close to the 4th and the 5th (Figure 7b).
response for every tower are shown and the modal fre-
quencies, and associated damping ratios, are highlighted.
From the EFDD results it is clearly remarkable 3.2.2. Stochastic subspace identification
(Figure 7) that the frequencies are in the range of The SSI method (Peeters and Roeck 1999), instead, is
1 ÷ 5 Hz. As may be expected, the Matildea tower is the applied in the time domain. It is the most used identi-
most rigid one (f1 = 1.472 Hz), due to its little height and fication technique among the OMA methodologies. It
its double box structure, whereas San Benedetto appeared mainly consists of the application of a mathematical
to be the most deformable tower (f1 = 0.762 Hz), since it model which upgrades its parameters in order to match
has only two floors that interrupt the elevation. At a first the raw time series data characteristics. It is based on
sight, it is observable that, apart for San Benedetto, all the the conversion of the second order problem, expressed
towers frequencies display the same trend: the first two by the dynamical equation, into a set of two equations,
frequencies are in the range 1 ÷ 1.5 Hz (Figure 7), with through the construction of State Space models: the
percentage differences around 3.1–2.9%, for San Giorgio “state equation” (3) and the “observation equation”
Belfry (Figure 7a) and Matildea Tower (Figure 7c), and (4), here expressed in their discretized form:
6.8% for Pomposa Abbey Belfry (Figure 7d): the proxi- f^xkþ1 g ¼ ½Af^xk g þ ½Bfuk g (3)
mity of the first two frequencies is typical of symmetrical
structures. The third frequency is in an intermediate
position among the first two ones and the 4th and 5th fyk g ¼ ½Cf^xk g þ ½Dfuk g (4)
ones, which also result very close to each other (with where
percentage difference always inferior to the 10%). The
differences presented by the tower of San Benedetto are ● ½A is the matrix of input physical information;
● ½B is the matrix of input statistical parameters; frequency is located in an intermediate position among
● ½C is the discrete output matrix; the first two ones and the 4th and 5th ones, which also
● ½D is the direct output transmission matrix; result very near between themselves (with percentage
● xk ¼ xk ðΔtÞ is the discrete-time vector containing difference always less than 10%).
the sampled displacement and velocities;
● fyk g; fuk g are the vectors of sampled input and
3.2.3. Comparison and validation of the
output.
experimental results
A first comparison between the results produced using
Other parameters describing the dynamic system are both methodologies was operated in terms of frequen-
considered as deterministic and should not be subdued cies, damping ratios and complexity. In particular, the
to the variations linked to excitation changes. comparison between the frequencies is expressed
The frequencies and damping ratios of the four struc- through the formula (Friswell and Mottershead 1995):
tures, extracted through the application of the SSI
method, are shown in Figure 8, where the Singular fi;SSI fi;EFDD
Δfi ½% ¼ 100 (5)
Value Decomposition (SVD) graphs are reported. fi;SSI
As seen before for the EFDD, in this case, all the
towers’ frequencies display the same trend, apart for where fi;SSI and fi;EFDD are the natural frequencies for the
San Benedetto: the first two frequencies are in the range ith mode, obtained with the SSI and EFDD methods. In
1 ÷ 1.5 Hz (Figure 8), typical of symmetric towers, with the following parts of the paper, in the comparisons
percentage differences around 2.7–2.8% for San Giorgio among the analytical and experimental frequencies, the
Belfry (Figure 8a) and Matildea Tower (Figure 8c), and subscript SSI will be assigned to the experimental ones,
8.3% for Pomposa Abbey Belfry (Figure 8), the third while the EFDD will correspond to the NM ones.
It became clear that both methods provided approxi- cases, except for San Benedetto in which the third modal
mately the same frequencies, in fact the differences are shape is not translational but flexural in Y-direction, as
lower than 2% except for the 2nd frequency of the San seen in the first mode. This difference may be traced back
Benedetto belfry where the difference is of ~5.3% to the inclination of the tower that is just this direction;
(Figure 10). The results that are more different instead the last two frequencies instead are flexural for all the
concern the complexity and the damping, as also the models. It is possible to observe a good level of correspon-
study of (Ramos et al. 2010) shows. dence between SSI and EFDD results.
It is notable that, even though the maximum differ- This feature is also confirmed through a validation
ence in terms of frequencies never exceeds the 5%, there process, operated applying MAC (Brincker, Zhang, and
is no coincidence between damping ratios extracted Andersen 2000; Ewins 1986), a tool used in case of
from EFDD and SSI, with these last ones that are always complex modes to execute a comparison between the
more elevate, except for the 2nd and 4th values of modal vectors describing a modal shape, providing an
Pomposa Belfry. This particular aspect can probably be indication about the consistency of a mode shape. It is
linked to the low intensity of the measured excitations, calculated as the normalized scalar product of the two
that do not allow to obtain reliable measures of the sets of vectors φA , which is the compatible analytical
damping ratios (Compan, Pachón, and Cámara 2017). modal vector associated to the r mode, and φX ,
The third comparison was operated on the mode which is the test modal vector associated to the q mode:
shapes associated to the first five modes analyzed for
each tower, which are reported in Figures 9–12. As stated T 2
φA φX q
in many studies on towers, it has been noted that the first MACðr; qÞ ¼
r
T T
(6)
two modes are usually translational while the third φA φA r ð φX φX q Þ
r q
appears to be rotational, as we may also notice in these
Figure 9. San Giorgio Belfry — Comparison between EFDD and SSI mode shapes.
12 G. STANDOLI ET AL.
Figure 10. San Benedetto Belfry — Comparison between EFDD and SSI mode shapes.
The scalar results are arranged in the MAC matrix, between test models and FEM models (Pastor, Binda,
whose optimal configuration is that of values near or equal and Tomáš 2012). CrossMAC matrices are reported
to 1 for the principal diagonal and equal to 0 outside of the in Tables 1–4, as can been seen the data oscillate
diagonal, assuring the perfect independency of the modes. between 63% and 99.80% a sign that the two methods
The MAC was used in this paper for the validation have identified very similar modes.
process of the modes selected with the EFDD and SSI Due to the good correspondence of the results
methods, and to assure the coherence of results between the two methodologies, the identification pro-
through the comparison between the results pro- cess was considered satisfactory thus we decided to use
duced by these different methodologies (Torres only the parameters extracted from the SSI approach as
et al. 2017). This last operation was conducted the ones for the definition of the experimental model
through a variation of the MAC criterion: the for the following calibration of the NMs.
CrossMAC. Like in the classical MAC criterion,
every possible combination of analytical and test
vector, associated to the ith mode are checked, but 4. Model updating process
the improved version of the consistency correspon-
4.1. Preliminary FE modeling
dence criterion is calculated using as sets of rows and
columns of the MAC matrix, modal shape vectors After obtaining the real dynamic behavior, the next step
coming from different sets of modes. This make the is represented by the construction of the NMs. In this
method particularly suitable for evaluate the affinity study the four bell-towers were analyzed using FEM,
INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 13
Figure 11. Matildea tower — Comparison between EFDD and SSI mode shapes.
that were created using the MidasFea© software modeled, due to their contribution on the stiffness.
(Figure 13).The NMs were carried out starting with The other (secondary) structural elements as the wood
the representations of their geometry, paying special floors were not modelled but were taken into account
attention to all the geometrical characteristics that by applying their dead loads on the walls. All the
may affect their dynamic behavior such as the open- belfries were considered fixed at the basis and isolated
ings, the walls thickness, etc. Afterward, the geometries since the thin walls that connect Pomposa and Matildea
were discretized using 4-nodes tetrahedral solid ele- Towers with the relative churches are irrelevant. These
ments, to whom were assigned, after a visual inspection towers were all built with masonry brick and lime
and historical analysis of the typology of masonries, the mortar except for the San Giorgio bell-tower where
material parameters reported in the Italian Code the bearing structure of bricks is covered by white
(Ministero delle Infrastrutture e dei Trasporti 2018). and rose calcareous stones.
In Table 5 are reported the number of elements, the Table 6 reports the elastic parameters assigned to NMs
number of nodes and the degrees of freedom for each at the early stage of this study; the values were taken
model. following the Italian Code considering the lowest
As may be observed in (Figure 13) the masonry cross Knowledge Level (KL1). These parameters according to
vaults, the masonry stairs and the wood roofs were the thickness of the mortar and conservation status were
14 G. STANDOLI ET AL.
Figure 12. Pomposa Abbey Belfry — Comparison between EFDD and SSI mode shapes.
corrected by applying the coefficients of the Table.C8.5.II Comparing the numerical with the experimental
of (Ministero delle infrastrutture e dei trasporti 2019). results, we can notice some differences especially in
The NMs were tested by modal analysis using the terms of frequencies; in Table 8 we reported the per-
Lanczos method to identify the dynamic behavior. centage errors between experimental and numerical
The first five modal shapes and the characteristic of frequencies. Those results confirm that visual analysis
their dynamic motion for each tower are reported in and the use of Code materials parameters are not
Figure 14 and Table 7. enough to obtain models that truly represent
INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 15
a building’s behaviour due to all the uncertainties that experimental evidences, while the modal shapes seem
there can be. to be well represented. For this purpose, the first opera-
tion related to the increase of the stiffness of the angu-
lar columns, that appear to be formed by squared stone
4.2. Calibration process blocks connected with a good quality mortar, and the
The preliminary results highlight the necessity of integration of rigid links at the level of the floors for
a recalibration of the NMs’ parameters (Diaferio et al. which the stiffness was not certain. As concerns the
2019), in order to have correspondence between experi- vaulted ceiling and the arcs, considering the construc-
mental and numerical models in the modal shapes tive techniques of the time, we assumed an increase of
associated with each vibrational mode. The modal the density to a value equal to 23 kN/m3 and a modulus
updating procedure was executed in an indirect way, of 2250 MPa. Those changes produced a raise of the
through the modification of the mechanical properties torsional component, associated with the third mode.
of masonry, varying the values of the (elastic) Young’s In the second step, we tried to increase the third mode
Modulus E and the density ρ of materials and paying shape. For this purpose, we augmented the elastic module
attention to the stiffness of the floors. We opted for an of the last nut, in order to differentiate it from the inferior
iterative procedure to manually update the parameters, part that may be less resistant due to the many changes
which consists in modifying the stiffness characteristics suffered over the years as seen in the Section 2.1. This
of the NM in order to make the frequencies coincident process also leads to obtain a better matching with the
from an engineering point of view. The process ran for bending modal shapes characterizing the 4th and 5th
each tower is illustrated below. mode. The final results are reported in Table 9 and
Figure 15.
Figure 13. Towers NMs: (a) prospective views; (b) section views.
assigned to the masonry constituting the higher block elastic Young’s Modulus equal to 1000 MPa, which
of the tower. highlights the severity of the damage suffered by the
In the second step, in order to represent in structure due to the seismic event of 2012. In
a realistic way the deterioration of the mechanical Table 11 and Figure 17 differences between the
properties of the structure (Cavalagli, Comanducci, numerical model and the experimental data are
and Ubertini 2018), we adopted for the bell cell an shown.
18 G. STANDOLI ET AL.
Table 8. Comparison between experimental and preliminary resulted less damaged during the preliminary visual
numerical models’ frequencies. inspection. The following operation has been the
San Giorgio S. Benedetto Matildea Pomposa assignment of different properties to the materials
Belfry Belfry Tower Belfry
in the X and Y direction with the aim of better
Mode Δf [%] Δf [%] Δf [%] Δf [%]
simulating the bigger degradation endured by the
1 6.41 0.39 50.97 9.38
2 8.87 41.57 48.58 12.99 East — West facade due to the seismic events and
3 17.93 9.53 37.72 11.04 atmospheric agents. At last, the Elastic Young's
4 18.15 7.44 15.46 13.24
5 10.99 4.11 16.20 5.66 Modulus of the top level of the tower has been
reduces, because of the worse quality of masonry
and the negative effect of lower normal stresses in
Table 9. San Giorgio Belfry — Comparison between experimen-
this part of the tower.
tal and calibrated NMs frequencies.
fEXP fNM Δf Mass Tran. Mass Tran. Mass Rot.
The modal shapes and the difference between the
Mode [Hz] [Hz] [%] X [%] Y [%] Z [%] numerical and experimental data, obtained at the
1 1.030 1.023 0.680 55.86 6.23 0.06 end of the last step, are reported in Table 12 and
2 1.060 1.061 0.090 6.16 55.71 0.00
3 3.213 3.056 4.890 0.00 0.00 5.12 Figure 18.
4 4.656 4.576 1.720 2.50 18.68 0.00
5 4.703 4.812 2.320 18.33 2.67 0.00
4.3. Validation of mode shape through CrossMAC
matrix
4.2.4. Pomposa Abbey belfry The model updating process seemed to produce a good
At a first glance, the exam of the results in terms of match between the experiment and NM, both in terms
frequencies displayed how difficult it is to differentiate of frequencies and modal shapes (Clementi et al. 2017;
the frequencies associated with the first two modes, and Pellegrini et al. 2018). To be sure of the outcome’s
the necessity to reduce the third one. reliability, the mode shapes produced by the NMs
The first step of the calibration (Elyamani et al. were compared with the corresponding ones deriving
2017) consisted in removing the conical cover of the from the modal identification executed through the SSI
tower. During the AVT, that part of the tower was not estimation methodology, using the Cross MAC
instrumented due to access difficulties, thus it was explained in §3.2. The results for each tower are pre-
considered a dead load in order to eliminate its influ- sented in Table 13.
ence on the first modes. This expedient permitted Even though the results are not satisfying, in par-
a better identification of the first two translational ticular from the fourth mode where the values are in
modes, and a significant increment of the frequency the ranges of 0.01 ÷ 0.3, we obtained an improve-
associated with the fourth one. ment of the results during the iterative step by step
The following step of the calibration aimed to process, which proofs that the adopted procedure is
increase the gap between the first two frequencies correct.
and to reduce the third one. In order to achieve The discrepancy of the results is probably related
this goal, the first step consisted in the increment to an inhomogeneity of the material’ properties, due
of the stiffness of the lower part of the tower, that to cracks, followed to the seismic events occurred in
INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 19
Figure 15. San Giorgio Belfry — Comparison between experimental and numerical mode shapes after calibration.
Table 10. San Benedetto belfry — Comparison between experi- the variations of the elastic modulus in the three
mental and calibrated NMs frequencies. spatial direction. This will be the subject of future
fEXP fNM Δf Mass Tran. Mass Tran. Mass Rot. works.
Mode [Hz] [Hz] [%] X [%] Y [%] Z [%]
1 0.759 0.737 2.900 0.00 53.33 0.00
2 1.380 1.372 0.580 52,75 0.00 0.00
3 2.781 2.866 3.060 0.44 19.35 0.03
4 2.858 2.868 0.350 19.48 0.45 0.00 5. Conclusions
5 2.982 2.994 0.400 0.02 0.04 0.00
The subjects of this paper are four historical masonry
belfries located in the North-East of Italy. They have
recent years, and to localized deterioration of the been subject at AVTs in order to identify the dynamics
connections among the bricks, forming the masonry characteristics such as the frequencies and modal
walls, whose effects would be probably better repre- shapes, to be able to characterize their NMs. Time
sented through the usage of anisotropic material histories recorded were cleansed before being analyzed
(Gentile, Saisi, and Borlenghi 2019), taking into using EFDD and SSI identification techniques, in order
account the influence of shear modulus (G) and to extract the dynamic characteristics of the structures.
20 G. STANDOLI ET AL.
Figure 16. San Benedetto Belfry — Comparison between experimental and numerical mode shapes after calibration.
Figure 17. Matildea Tower — Comparison between experimental and numerical mode shapes after calibration.
Table 12. Pomposa Abbey Belfry — Comparison between CrossMAC between the numerical and experimental
experimental and calibrated NMs frequencies. modes at the end of the calibration. The results show
fEXP fNM Δf Mass Tran. Mass Tran. Mass Rot. both good values for the first three frequencies and
Mode [Hz] [Hz] [%] X [%] Y [%] Z [%]
1 0.939 0.934 0.530 45.67 0.06 0.00
a significant difference for higher modes that typi-
2 1.024 1.018 0.590 0.05 45.57 0.00 cally ascribable to very strong anisotropy behavior of
3 2.754 2.818 2.320 0.02 1.31 0.04
4 3.493 3.433 1.720 23.62 0.02 0.00
masonries.
5 3.516 3.487 0.820 0.03 24.11 0.00 The results of these procedures emphasize once
again that the knowledge of historical structures based
merely visual surveys and historical analysis is not
comparison; in fact, being these differences always enough to reproduce a reliable NM. For this reason,
lower than the 5% comparing numerical and experi- more accurate analyses are necessary, such as the non-
mental models. To the contrary, more steps were destructive ones. The AVT followed by the model
necessary to gain a better coincidence between updating appears to provide more positive results as
mode shapes. For this reason, we analyzed the a starting point to carry out more extensive analysis.
22 G. STANDOLI ET AL.
Figure 18. Pomposa Abbey Belfry — Comparison between experimental and numerical mode shapes after calibration.
Table 13. Cross-MAC between experimental and numerical mode shapes after calibration.
San Giorgio Belfry San Benedetto Belfry
Mode fEXP [Hz] fNM [Hz] Δf [%] CrossMAC [%] Mode fEXP [Hz] fNM [Hz] Δf [%] CrossMAC [%]
1 1.030 1.023 0.68 89.70 1 0.759 0.737 2.90 99.20
2 1.060 1.061 0.09 86.30 2 1.38 1.372 0.58 63.80
3 3.213 3.056 4.89 100.00 3 2.781 2.866 3.06 4.80
4 4.656 4.576 1.72 68.30 4 2.858 2.868 0.35 9.60
5 4.703 4.812 2.32 66.70 5 2.982 2.994 0.40 14.90
Matildea Tower Belfry Pomposa Abbey Belfry
Mode fEXP [Hz] fNM [Hz] Δf [%] CrossMAC [%] Mode fEXP [Hz] fNM [Hz] Δf [%] CrossMAC [%]
1 1.478 1.478 0.00 84.70 1 0.939 0.934 0.53 73.50
2 1.513 1.511 0.13 97.20 2 1.024 1.018 0.59 96.30
3 4.206 3.917 6.87 38.40 3 2.754 2.818 2.32 72.80
4 5.011 4.767 4.87 0.50 4 3.493 3.433 1.72 5.40
5 5.14 5.169 0.56 35.20 5 3.516 3.487 0.82 1.50
INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 23
Facchini, L., and M. Betti. 2016. Simplified seismic analysis of example. Journal of Civil Structural Health Monitoring
disordered masonry towers. ASCE-ASME Journal of Risk and (December). doi:10.1007/s13349-019-00368-1.
Uncertainty in Engineering Systems, Part A: Civil Engineering Lenci, S., L. Consolini, and F. Clementi. 2015. On the experi-
2 (2):C4015010. doi:10.1061/AJRUA6.0000856. mental determination of dynamical properties of lami-
Ferrante, A., F. Clementi, and G. Milani. 2019. Dynamic nated glass. Annals of Solid and Structural Mechanics 7
behavior of an inclined existing masonry tower in Italy. (1–2):27–43. doi:10.1007/s12356-015-0040-z.
Frontiers in Built Environment 5:33. doi:10.3389/fbuil.20 Lopez, S., M. D’Amato, L. Ramos, M. Laterza, and
19.00033. P. B. Lourenço. 2019. Simplified formulations for estimat-
Foti, D., M. Diaferio, N. I. Giannoccaro, and M. Mongelli. ing the main frequencies of ancient masonry churches.
2012. Ambient vibration testing, dynamic identification Frontiers in Built Environment 5 (March). doi: 10.3389/
and model updating of a historic tower. NDT & fbuil.2019.00018.
E International 47 (April):88–95. doi:10.1016/j.ndteint.20 Milani, G. 2013. Lesson learned after the Emilia-Romagna,
11.11.009. Italy, 20–29 May 2012 earthquakes: A limit analysis insight
Friswell, M. I., and J. E. Mottershead. 1995. Finite element on three masonry churches. Engineering Failure Analysis
model updating in structural dynamics. Vol. 38 Solid 34:761–78. doi:10.1016/j.engfailanal.2013.01.001.
Mechanics and Its Applications. Dordrecht: Springer Milani, G., R. Shehu, and M. Valente. 2017. Role of inclination
Netherlands. doi:10.1007/978-94-015-8508-8. in the seismic vulnerability of bell towers: FE models and
Fuentes, D. D., P. A. B. Julià, M. D’Amato, and M. Laterza. simplified approaches. Bulletin of Earthquake Engineering
2019. Preliminary seismic damage assessment of mexican 15 (4):1707–37. doi:10.1007/s10518-016-0043-0.
churches after September 2017 earthquakes. International Ministero delle infrastrutture e dei trasporti. 2019. Circolare
Journal of Architectural Heritage 1–21. doi:10.1080/ 21 Gennaio 2019 n. 7 C.S.LL.PP. Istruzioni per l’applica-
15583058.2019.1628323. zione Dell’aggiornamento Delle ‘Norme Tecniche per Le
Gade, S., N. B. Møller, H. Herlufsen, and H. Konstantin- Costruzioni’ Di Cui Al D.M. 17/ 01/2018 (in Italian).
Hansen. 2005. Frequency domain techniques for opera- Suppl. Ord. Alla G.U. n. 35 Del 11/2/19.
tional modal analysis. First International Operational Ministero delle Infrastrutture e dei Trasporti. 2018.
Modal Analysis Conference, no. Type 7760. 261–71. Aggiornamento Delle ‘Norme Tecniche per Le
Copenhagen. Costruzioni’ - NTC 2018 (in Italian). 1–198.
Gentile, C., and A. Saisi. 2007. Ambient vibration testing of Pastor, M., M. Binda, and H. Tomáš 2012. Modal assurance
historic masonry towers for structural identification and criterion. In Procedia Engineering. doi:10.1016/j.proeng.
damage assessment. Construction and Building Materials 2012.09.551. Zemplínska Šírava, Slovakia.
21 (6):1311–21. doi:10.1016/j.conbuildmat.2006.01.007. Peeters, B., and G. D. Roeck. 1999. Reference-based stochastic
Gentile, C., A. Saisi, and P. Borlenghi. 2019. Fe modelling for subspace identification for output-only modal analysis.
seismic assessment of an ancient tower from ambient Mechanical Systems and Signal Processing 13 (6):855–78.
vibration. In IOMAC’19–8th International Operational doi:10.1006/mssp.1999.1249.
Modal Analysis Conference, 1–11. Copenaghen. Pellegrinelli, A., A. Furini, and P. Russo. 2014. Earthquakes
Gentile, C., M. Guidobaldi, and A. Saisi. 2016. One-year and ancient leaning towers: Geodetic monitoring of the
dynamic monitoring of a historic tower: Damage detection bell tower of San Benedetto church in Ferrara (Italy).
under changing environment. Meccanica 51 (11):2873–89. Journal of Cultural Heritage 15 (6):687–91. doi:10.1016/j.
doi:10.1007/s11012-016-0482-3. culher.2013.12.005.
Invernizzi, S., G. Lacidogna, N. E. Lozano-Ramírez, and Pellegrini, D., M. Girardi, P. B. Lourenço, M. G. Masciotta,
A. Carpinteri. 2019. Structural monitoring and assessment N. Mendes, C. Padovani, and L. F. Ramos. 2018. Modal
of an ancient masonry tower. Engineering Fracture analysis of historical masonry structures: Linear perturba-
Mechanics 210 (April):429–43. doi:10.1016/j.engfracmech. tion and software benchmarking. Construction and
2018.05.011. Building Materials 189 (November):1232–50. doi:10.1016/
Kita, A., N. Cavalagli, and F. Ubertini. 2019. Temperature j.conbuildmat.2018.09.034.
effects on static and dynamic behavior of Consoli Palace in Pierdicca, A., F. Clementi, A. Fortunati, and S. Lenci. 2019.
Gubbio, Italy. Mechanical Systems and Signal Processing Tracking modal parameters evolution of a school building
120 (April):180–202. doi:10.1016/j.ymssp.2018.10.021. during retrofitting works. Bulletin of Earthquake
Korumaz, M., M. Betti, A. Conti, G. Tucci, G. Bartoli, Engineering 17 (2):1029–52. doi:10.1007/s10518-018-0483-9.
V. Bonora, A. G. Korumaz, and L. Fiorini. 2017. An Preciado, A. 2015. Seismic vulnerability and failure modes
integrated terrestrial laser scanner (TLS), deviation analysis simulation of ancient masonry towers by validated virtual
(DA) and finite element (FE) Approach for health assess- finite element models. Engineering Failure Analysis 57
ment of historical structures. A minaret case study. (November):72–87. doi:10.1016/j.engfailanal.2015.07.030.
Engineering Structures 153 (July):224–38. doi:10.1016/j. Quagliarini, E., G. Maracchini, and F. Clementi. 2017. Uses
engstruct.2017.10.026. and limits of the equivalent frame model on existing
Lagomarsino, S., and S. Podesta`. 2004. Damage and vulner- unreinforced masonry buildings for assessing their seismic
ability assessment of churches after the 2002 Molise, Italy, risk: A review. Journal of Building Engineering 10
earthquake. Earthquake Spectra 20 (S1):S271–83. doi:10.11 (March):166–82. doi:10.1016/j.jobe.2017.03.004.
93/1.1767161. Ramos, L. F., L. Marques, P. B. Lourenço, G. De Roeck,
Laura, M., C. Francesco, and F. Antonio. 2019. Static and A. Campos-Costa, and J. Roque. 2010. Monitoring historical
dynamic testing of highway bridges: A best practice masonry structures with operational modal analysis: Two
INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 25
case studies. Mechanical Systems and Signal Processing 24 metropolitan cathedral of Santiago, Chile. Engineering
(5):1291–305. doi:10.1016/j.ymssp.2010.01.011. Structures 143:169–88. doi:10.1016/j.engstruct.2017.04.008.
Rodrigues, J. 2004. Identificação Modal Estocástica: Métodos de Ubertini, F., G. Comanducci, and N. Cavalagli. 2016.
Análise e Aplicações Em Estruturas de Engenharia Civil. Vibration-based structural health monitoring of
Faculdade de Engenharia Da Universidade Do Porto, Porto. a Historic Bell-Tower using output-only measurements
Shakya, M., H. Varum, R. Vicente, and A. Costa. 2016. Empirical and multivariate statistical analysis. Structural Health
formulation for estimating the fundamental frequency of slen- Monitoring: An International Journal 15 (4):438–57.
der masonry structures. International Journal of Architectural doi:10.1177/1475921716643948.
Heritage 10 (1):55–66. doi:10.1080/15583058.2014.951796. Ubertini, F., G. Comanducci, N. Cavalagli, A. L. Pisello,
Stefano, A. D., E. Matta, and P. Clemente. 2016. Structural A. L. Materazzi, and F. Cotana. 2017. Environmental
health monitoring of historical heritage in Italy: Some effects on natural frequencies of the San Pietro Bell
relevant experiences. Journal of Civil Structural Health Tower in Perugia, Italy, and their removal for structural
Monitoring 6 (1):83–106. doi:10.1007/s13349-016-0154-y. performance assessment. Mechanical Systems and Signal
Torres, W., J. L. Almazán, C. Sandoval, and R. Boroschek. 2017. Processing 82 (January):307–22. doi:10.1016/j.ymssp.2016.
Operational modal analysis and FE model updating of the 05.025.