Engineering Structures 123 (2016) 354–371

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Calibration of the numerical model of a stone masonry railway bridge

based on experimentally identified modal parameters
C. Costa a,⇑, D. Ribeiro b, P. Jorge c, R. Silva c, A. Arêde c, R. Calçada c
a
CONSTRUCT-LESE, Polytechnic Institute of Tomar (IPT), Tomar, Portugal
b
CONSTRUCT-LESE, School of Engineering, Polytechnic of Porto, Porto, Portugal
c
CONSTRUCT-LESE, Faculty of Engineering (FEUP), University of Porto, Porto, Portugal

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

Article history: This paper focuses on the calibration of a numerical model of a stone masonry arch railway bridge using
Received 16 November 2015 dynamic modal parameters estimated from an ambient vibration test. The developed 3D numerical
Revised 22 May 2016 model is based on the finite element method, featuring a realistic representation of the bridge structural
Accepted 25 May 2016
components and materials. The calibration methodology relied on a genetic algorithm strategy which
allowed estimating and updating numerical model parameters, particularly the elastic properties of
materials. The validation of the updated bridge material properties’ parameters was based on the results
Keywords:
of material testing, on existing bridge design data and on observations resulting from in situ visual
Stone masonry
Railway bridges
inspections.
Numerical modelling Ó 2016 Elsevier Ltd. All rights reserved.
Vibration testing
Operational modal analysis
Genetic algorithm

1. Introduction
The need to identify exploitation limits and conditioning effects
of regular traffic operation on stone masonry arch railway bridges
led to the development of the StonArcRail research project.
This study comprised experimental and numerical research
activities on the effects of rail traffic in the structural behaviour
of bridges. It includes identifying the vibration effects caused by
traffic action and the influence of its parameters (speed, type of
train and track irregularities) on the dynamic response of the
bridge, track and train subsystems, aiming at assessing the struc-
ture and the track safety as well as the passenger comfort.
The large number of existing and in operation masonry arch
bridges on the rail network across Europe justifies the need to
study this type of bridges. According to data reported in UIC [1],
which several European have contributed to, about 60% of railway
bridges are arched ones or culverts. In Portugal there are about
11 746 such cases which amount to 90% of the total existing rail-
way bridges. The report also concludes that 80% of these Por-
tuguese bridges have spans lower than 5 m and 70% are aged
between 100 and 150 years. The European project ‘‘Sustainable
⇑ Corresponding author.
E-mail addresses: c.costa@ipt.pt (C. Costa), drr@isep.ipp.pt (D. Ribeiro),
pfmjorge@fe.up.pt (P. Jorge), rubenfsilva@fe.up.pt (R. Silva), aarede@fe.up.pt
(A. Arêde), ruiabc@fe.up.pt (R. Calçada).
Bridges” [2] which involved about 220 000 bridges in Europe, also
concluded that about 41% are arched bridges, of which 35% are
over 100 years old and 62% have small span.
Due to their age, the characterization of these bridges’ conser-
vation status is essential to allow assessing their structural beha-
viour and identifying the exploitation limits for future rail
networks. For this purpose, experimental campaigns on these
bridges should be made, mainly involving non-destructive testing,
as presented by McCann and Forde [3] in a review of such type of
experimental methods applicable to and used on concrete and
masonry structures. Orbán and Gutermann [4], refer the UIC pro-
ject results relative to methods of inspection and testing of railway
bridges in arched masonry. Also, Olofsson et al. [2], present the
results of the European project ‘‘Sustainable Bridges” describing
methods to upgrade existing railway bridges in the European net-
work. Roberts and Hughes [5] introduce a monitoring task through
the use of accelerometers in the upper part of the superstructure.
Brencich and Sabia [6,7] investigated masonry bridges both in ser-
vice conditions and in various demolition stages, having developed
an experimental program involving lab tests on material samples,
as well as in situ activities, namely flat jack and sonic tests to char-
acterize material properties and dynamic tests for the identifica-
tion of modal shapes and frequencies. Arêde et al. [8] presented
the results of a project which involved the implementation of an
instrumentation system on a stone arch bridge during the con-
struction (2004–2005) and its structural response monitoring

http://dx.doi.org/10.1016/j.engstruct.2016.05.044
0141-0296/Ó 2016 Elsevier Ltd. All rights reserved.
 C. Costa et al. / Engineering Structures 123 (2016) 354–371
during a load test. This project also included ambient vibration [9]
and lab testing of the materials used in the bridge construction,
namely classic tests for material strength characterization in stone
specimens and mortar, triaxial tests on filling material samples and
shear-compression test on samples representative of the joints
[10]. Srinivas et al. [11] present experimental methodologies using
flat jacks to assess the longitudinal internal forces on the bridge
due to the increased weight of freight wagons.
The evaluation of structural operation and load capacity of
masonry arch bridges always lacked appropriate numerical tools
for their analyses. The first scientific approaches were based on
the concepts of static and failure mechanisms experimentally
observed, such as the works of Philippe de la Hire [12], Pippard
[13] and Heyman [14]. Later on, thanks to the rapid evolution of
computational technologies, new methodologies based on the
finite element method (FEM) and the discrete elements methods
(DEM) are developed and increasingly used, some of which are
addressed in the following paragraphs.
Fanning et al. [15] used a 3D FEM model in which masonry, fill-
ing and pavement were discretized with continuous solid elements
using commercial software ANSYS [16]. Frunzio et al. [17] studied a
Roman arch bridge adopting a 3D FEM model developed also in
ANSYS, wherein the nonlinear material behaviour was considered
using the Drucker-Prager criterion for all materials. Cavicchi and
Gambarotta [18] used a 2D FEM where the arch and backfill inter-
action was considered, the arches and piers were modelled with 1D
elements with perfect elasto-plastic material behaviour, ductile in
compression and with no tensile strength. The backfill was simu-
lated by 2D triangular finite elements interconnected by interface
elements ruled by a modified Mohr-Coulomb criterion with ten-
sion ‘‘cut-off”. Arani and Zandi [19] also studied the conditions of
a three span railway bridge, using a FEM based model with
SOLID-2D plain strain elements. Detailed FEM modelling strategies
were adopted by Costa et al. [20] for simulating the structural
response of stone arch bridges under road traffic loading, resorting
to the computer code CAST3M [21]. In the adopted models the
bridges’ masonry components are represented by FEM micro mod-
elling strategies using solid elements, to define the individualized
blocks, and zero thickness joint elements at their interfaces. The
backfill is also modelled with solid elements connected to zero
thickness joint elements in the interfaces between the infill and
blocks of the masonry structure. Nonlinear constitutive models
supported by experimental calibration were considered assuming
a nonlinear Mohr-Coulomb friction model without dilatancy for
the joint elements and using the Drucker-Prager model for the
infill material. In the sequence of this work, DEM models were also
used for comparative purposes to evaluate the load-carrying
capacity of one of the case studies [22]. Anderson [23] developed
a comprehensive 2D model of a railway concrete arch bridge and
a 3D model of the three spans on the north side of the bridge, using
the SOLVIA03 program [24]. For the structural modelling, the con-
crete was considered as a continuous nonlinear distributed crack-
ing material and the backfill behaving the Drucker-Prager criterion.
The 3D model involved the separation between the various con-
stituent elements, namely arches, spandrel walls, backfill, piers
and foundations. Domede et al. [25] studied an arched masonry
railway bridge using a 3D FEM based damage model developed
in CAST3M [21]. The masonry was simulated as a nonlinear
homogenized solid material by means of a damage model in which
the masonry behaviour in compression and tension is handled sep-
arately. The backfill was modelled with a nonlinear homogeneous
material following the Drucker-Prager criterion.
Comparing the numerical results with the experimental ones, it
is normal to have differences that should be minimized by optimiz-
ing the bridge numerical models [26]. Bayraktar et al. [27,28] pre-
sent the numerical modelling and calibration of roadway stone
355
masonry arch bridges for which people walking was adopted as
the dynamic excitation during vibration testing; in this case the
numerical model calibration only involved changes of the bridge
boundary conditions.
In line with previous studies, this paper presents the outcome of
a very recent work involving 3D FEM modelling of a stone masonry
arch railway bridge over 100 years old. The model was generated
in ANSYS [16], with refined individualization of the various bridge
components in order to allow assigning different material param-
eters to each of them. These parameters were considered on the
basis of an extensive experimental campaign involving in situ
and laboratory tests. Aiming at assessing the bridge response,
ambient vibration tests were performed from which natural fre-
quencies, vibration modes and damping coefficients were
obtained. A good correlation was obtained between the numerical
and experimental results, for which an optimization procedure was
adopted to improve the numerical model based on data obtained
from the ambient vibration tests. The calibrated model was devel-
oped to allow performing analyses involving the bridge-train inter-
action, from which the response of both can be obtained.
2. Durrães railway bridge
2.1. Description
The Durrães bridge (Fig. 1) dates back to late 19th century and
is located at km + 64.344 of the Minho line which constitutes the
rail link between the Porto and Valença cities. The bridge presents
a structural system made of granite masonry arches and is part of a
single-track section in Iberian gauge currently allowing the circula-
tion of freight and passenger trains with maximum speeds of
100 km/h and 120 km/h, respectively.
The bridge length is about 178 m, having a longitudinal rectilin-
ear profile deck with 1.45% slope and 5.3 m width. It consists of
16 arches with approximately 9 m span, supported by 15 piers
and two abutments. The spandrel walls, vertically supported on
the outer faces of arches and piers, are formed by horizontal rows
of carved stone. The maximum gap between the bridge ground
level and the railroad is approximately 22 m. Fig. 2 shows the west
view of Durrães Bridge obtained from a topographical survey made
within the StonArcRail project activities.
The arches have 0.7 m uniform thickness and the piers’ height,
measured between the top face of the foundations’ blocks and the
arches’ bases, range from 11 m to 12 m. The two piers located at
about 1/3 and 2/3 of the total length of the bridge (between arches
A5 and A6 and arches A11 and A12) have about twice the cross sec-
tion area of the other piers. The shallow foundations are based on a
rocky base at a variable depth between 5 m and 10 m.
The railway line consists of bi-block type sleepers and UIC60
rails, laid on a ballast layer approximately 0.50 m thick. The side
guards are made of granite stone blocks.
2.2. Preliminary experimental campaign
An experimental campaign was carried out aiming at studying
and evaluating the physical and mechanical parameters of the
structural components of two stone masonry bridges, including
the herein addressed Durrães bridge, described at length (including
presentation of results) in another paper prepared by some of the
present paper authors [29].
The characterization of the Durrães bridge materials consisted
of in situ tests, namely, flat jack (FJ) and Ménard Pressuremeter
(PMT) tests, and laboratory tests on samples of stone blocks and
joints extracted from the bridge. Tests with Ground Penetrating
Radar (GPR) were also carried out in order to study the geometric
356 C. Costa et al. / Engineering Structures 123 (2016) 354–371

Fig. 1. Durrães bridge.

Fig. 2. Durrães bridge – longitudinal West view.

constitution of the sections of the structural elements and the ter-
rain’s foundation profile. For the latter purpose, Dynamic Probing
Super Heavy tests (DPSH) were also made. In total two FJ tests
were performed in some piers’ facing stones, as well as four PMT
tests (in two piers and two spandrel walls) and four DPSH tests
on four ground locations between piers. Fig. 3 shows the marked
locations of these tests. GPR tests were performed to obtain GPR
profiles from outer facings of the bridge located in abutments, piers
and spandrel walls and along the terrain profile between P1 and
P15 piers.
2.2.1. GPR (Ground Penetrating Radar) and DPSH tests
Radargrams’ analyses obtained with the GPR tests allowed iden-
tifying the thickness of facing stones and the position of foundation
soil layers. GPR tests were performed using a Malå GeoScience AB
equipment which included a control and registration unit, a power
source and two antennas with 250 and 500 MHz of frequency. To
perform the radargrams’ analyses the following computational
programs have been used: GroundVision (MALÅ GeoScience AB)
and Reflex2DQuick (Sandmeier geophysical software) to read and
treat the radargrams obtained in situ and GPRSIM and GPRSLICE
(Geophysical Archaeometry Laboratory) to simulate geometric
models. Using GPRSIM simulation software, synthetic radargrams
have been generated based on the geometry and dielectric con-
stants of the materials involved which revealed to be consistent
with the radargrams obtained in situ. Fig. 4a shows a measuring
phase of a pier profile and Fig. 4b includes the respective radar-
gram. Good agreement was obtained between the results esti-
mated based on the interpretation and processing of the
measured radargrams and other geometric data available, namely,
from the design drawings, DPSH tests and observation of layers of
masonry and infill materials extracted from drilled cores in the
bridge [29].
The determination of the firm depth was verified with the
results of the DPSH tests. In Fig. 5a it can be seen an overview of
the used equipment, while Fig. 5b shows the in depth plot evolu-
tion of the number of blows (N20) and dynamic resistance of the
tip (qd) obtained from the DPSH ground tests [29]. By observing
the evolution of N20 in the various DPSH tests, it was found the
presence of a firm layer at about 10 m depth in most of the bridge
length, where the piers’ foundations are sit on. The DPSH4 test
results show the presence of a rocky outcrop near the South abut-
ment, which raises the possibility of lower depth foundations at
that level.

DPSH1 to DPSH4 tests PMT1 to PMT4 Pressurmeter tests FJ1 to FJ2 Flat-jack tests
Fig. 3. Locations of the in situ experimental tests in the Durrães bridge.
 C. Costa et al. / Engineering Structures 123 (2016) 354–371 357

Fig. 4. GPR test: (a) measurements with 500 MHz antenna and (b) radargram.

Fig. 5. DPSH test: (a) equipment used for DPSH tests and (b) in depth evolution of the number of blows (N20).

Fig. 6. In situ FJ2 deformability test: (a) test setup and (b) experimental curves.

2.2.2. In situ flat-jack testing (FJ) These tests were performed in masonry facing stones, in accor-
In situ stress state and deformability properties of masonry dance with the test procedure described in [30] and the applicable
were estimated by single and double flat-jacks tests, respectively. standards [31,32]. Fig. 6a shows one of the double tests performed
358 C. Costa et al. / Engineering Structures 123 (2016) 354–371

and the layout of displacement measuring devices (LVDTs), four Table 1

vertical and one horizontal. Fig. 6b shows a graph with the load- Physical and mechanical parameters of granite stone blocks and masonry joints.

ing/unloading cycles obtained from a double test [29], from which Material Parameter Value Standard
it is possible to estimate elastic modulus of that masonry zone Granite stone Unit weight (kN/m ) 3
25.9–26.5 –
(consisting of stone blocks and joints) by calculating the average blocks Compressive strength 34.8–59.4 NP EN 12504-1
slope of the stress-strain plot in each vertical alignment. NP EN 12390-3
For the installed in situ stress, a value of 1.5 MPa was estimated, Tensile strength by diametric 3.7–5.4 NP EN 12390-6
compression (MPa)
resulting from the analysis of the results of single FJ tests, while the Elastic modulus (GPa) 20.0–23.5 NP EN 14580
masonry deformability modulus was found in the range 7–23 GPa
Joints Normal stiffness (MPa/mm) 1.652 –
which results from double FJ tests. A large variability in results was Shear stiffness (MPa/mm) 0.728 –
found and the upper limit value 23 GPa is clearly overestimated,
for which a very plausible reason is the type of masonry found in
the Durrães bridge piers with very large regular stones (0.5 m high
masonry joints taken from the bridge and the obtained parameters,
and more than 1 m long), which does not fit in the applicable stan-
in terms of shear and normal stiffness, are also included in Table 1.
dards for this type of tests.

2.2.5. Visual inspection

2.2.3. Pressuremeter tests (PMT)
A visual inspection to Durrães bridge was carried out aiming at
Ménard pressuremeter tests (Fig. 7a) allowed estimating the
identifying damage and defects. According to the observations
deformability characteristics of the backfill. The tests were per-
made, two types of defects were differentiated: defects related to
formed according to the French standard NF-P94-110 [33]. Based
environmental, physical and chemical actions and defects related
on pressuremeter curves obtained from testing it was possible to
to mechanical actions. The first feature in a wider form in the
calculate the deformation parameters designated by pressureme-
structure and, in general, are independent of the structural beha-
ter deformation modulus (GPMT) and pressuremeter modulus
viour involving problems associated with the presence of water,
(EPMT). Fig. 7b presents a pressuremeter curve regarding a test on
biological pollution and erosion as shown in Fig. 8.
a pier [29].
It was observed that in general the bridge facing stones have
For each test it was determined the range of variation between
water flowing, efflorescence, some stalactites, black films, vegeta-
the minimum and maximum value of the pressuremeter modulus
tion, moss and lichens. In areas not affected by the presence of veg-
(EPMT). For the tests PMT1 and PMT2, in pier P11, the range of val-
etation it is common to see block erosion. The lack of mortar in the
ues was between 450 and 680 MPa, for PMT3 test, on the spandrel
joints and the existence of open joints is more visible in the lateral
wall, the range was 280 the 400 MPa and for the PMT4 test, in the
west surface of the bridge and at higher levels. In blocks of the
abutment, the interval was between 190 and 480 MPa. The elastic
piers’ faces closer to the ground, land deposits, moss and humidity
modulus (E) of the backfill material can be estimated by dividing
are observed.
the pressuremeter modulus (EPMT) by an appropriate coefficient
The defects related to mechanical actions are associated with
as described in Briaud [34], for which it was adopted the 0.5 value,
the presence of cracks and deformations resulting from the struc-
often adopted for unweathered rock material in the foundation
tural behaviour due to static or dynamic loading. There are prob-
design framework.
lems such as: (i) longitudinal crack in the intrados of the arches
alignments underneath the spandrel walls, (ii) longitudinal crack-
2.2.4. Laboratory tests ing on the arches intrados along the central axis of the bridge,
The mechanical characterization of the stone extracted from the (iii) block fracture (iv) cracking at the joints on the interface
bridge was based on laboratory tests for determining the compres- between the mortar and blocks or within the mortar and (v) joint
sive and tensile strength as well as the elastic modulus in compres- opening. In Fig. 9 some examples of such anomalies are shown.
sion. Table 1 presents the results of these tests. The parameters In arches A3, A5, A10, A12, A13, A15 and A16 it can be observed
were evaluated according to the rules applicable to each type of longitudinal cracking on the arch intrados near the spandrel walls
test with the reference standard included in Table 1. Laboratory (east and/or west) and in some cases also in the axis of the bridge.
shear and compression tests were also made in samples of Occasionally cracking is observed in the mortar and/or in the

Fig. 7. PMT test: (a) Ménard pressuremeter and (b) experimental pressiometric curve.
 C. Costa et al. / Engineering Structures 123 (2016) 354–371 359

(a) Water flowing, efflorescence, (b) Vegetation and black films on the
stalactites and vegetation, on the west face on the connections between (c) Moss and lichens on a pier.
arch intrados. the top cornice and the spandrel wall.

(d) Block erosion on the connections (e) Rising damp; dirt deposits and moss
between the top cornice and the in piers’ facing blocks close to the (f) Lack of mortar in pier joints.
spandrel wall. ground.
Fig. 8. Defects associated with environmental, physical and chemical actions.

(a) Longitudinal cracking in the intrados of the arch.

(b) Block fracture. (c) Cracking at the joints. (d) Joint opening.
Fig. 9. Defects associated with mechanical actions.

connection between the mortar and the blocks, noticing that the 2.3. Numerical modelling
joints most affected by this anomaly are located generally at higher
levels along the arch crown. Fig. 10 shows the location of some of The numerical modelling of Durrães bridge was carried out
the observed defects, in both east and west elevations. using a 3D FEM model developed in the ANSYS program [16].
360 C. Costa et al. / Engineering Structures 123 (2016) 354–371

Fig. 10. Anomalies on the bridge: (a) west side; (b) east side.

Fig. 11 shows the numerical model of Durrães bridge with the
identification of the various structural components. This model
was based on a simplified initial model previously presented in
another publication [35]. In the final model the geometry of the
spandrel wall and the depth of the foundations were updated from
the initial model.
The arches, piers, spandrel walls, backfill, abutments, founda-
tions, embankment, ballast and sleepers, were modelled using vol-
umetric finite elements of triangular prismatic and parallelepiped
shape with 6 and 8 nodes, respectively. The rails were modelled
using beam finite elements. The side guards were modelled
through mass elements distributed over both spandrel walls. The
model’s geometry was based on a topographic survey carried out
on the bridge, as well as data from the experimental campaign,
including the tests with GPR and DPSH, and details available in
design drawings.
The boundary conditions were established using rigid supports
to block all displacements of the nodes of the finite element mesh
located at the base of foundations and abutments.
Table 2 outlines the geometric and mechanical parameters of
the numerical model of the bridge, including the designation, the
adopted value in the initial model and the respective unit. In addi-

1. Abutments
2. Foundations
3. Base of the piers A16
4. Piers
5. Arches A15
6. Spandrel walls A14
7. Lower backfill
8. Upper backfill A13 9
9. Embankment A12
10. Ballast
11. Sleepers A11
12. Rail A10
A9
A8 12
A7 11
10 5
A6
A5 8
A4
7 4
A3
A2 6
A1 3

1 2

Fig. 11. 3D numerical model of Durrães bridge.

 C. Costa et al. / Engineering Structures 123 (2016) 354–371 361

Table 2
Characterization of the parameters of the numerical model of the Durrães bridge.

Parameter Designation Initial value Exp. campaign values Units

Low. Up.
Efund Elastic modulus of masonry foundations 15.0 – – GPa
qfund Volumetric mass of masonry foundations 2497.5 – – kg/m3
Eg Elastic modulus of the masonry of arch, spandrel walls, 10.0 7 23 GPa
piers and abutments
qg Volumetric mass of the masonry of arch, spandrel walls, 2497.5 2396.3 2525.5 kg/m3
piers and abutments
Ee,inf Elastic modulus of the backfill Lower zone 1.00 0.980 1.470 GPa
Ee,sup Upper zone 0.60 0.416 1.026 GPa
qe,inf Volumetric mass of the backfill Lower zone 2303.8 2302.7 2438.5 kg/m3
qe,sup Upper zone 2201.8 2199.2 2378.2 kg/m3
Eat Elastic modulus of the embankment 0.75 – – GPa
qat Mass density of the embankment 2000 – – kg/m3
Eb Elastic modulus of ballast 0.145 – – GPa
qb Volumetric mass of ballast 2000 – – kg/m3
Et Elastic modulus of concrete (sleepers) 36.0 – – GPa
qt Volumetric mass of concrete (sleepers) 2890 – – kg/m3
Ea Elastic modulus of steel (rail) 210.0 – – GPa
qa Volumetric mass of steel (rail) 7850 – – kg/m3

tion there are also indicated the lower and upper limits of the iden-
tified parameters in the experimental campaign presented in Sec-
tion 2.2. It should be noted that there was a slight delay between
the definition of the initial model and the processing of data from
the tests, which is why the parameter values of the materials
adopted in the initial model do not all strictly coincide with the
average values obtained in the tests.
2.4. Modal parameters
Fig. 12 presents the frequency values and the corresponding
modal shapes of some of the bridge vibration modes obtained from
the numerical model developed based on the adopted values of the
parameters listed in Table 2.
Modes 1–3, 5–7 and 9 involve transverse bending of the piers
which, by structural compatibility, cause transverse movements
of the deck. Modes 4 and 8 involve the longitudinal bending of
the piers. The vibration modes shown are those that will be used
later in the calibration phase of the numerical model of the bridge.
3. Ambient vibration test
The ambient vibration test of Durrães bridge aimed at the
experimental identification of its modal parameters, namely, the
natural vibration frequencies, mode shapes and damping
coefficients.
3.1. Measurement setup
This test was performed based on a technique that considers
fixed reference points and mobile measurement points, where
there were used 20 PCB piezoelectric accelerometers, model
393B12, with a sensitivity equal to 10 V/g, a range of measurement
of ±0.5 g, a broadband resolution equal to 6.0 lg rms and a fre-
quency range between 0.15 Hz and 1000 Hz. The ambient vibration
response was evaluated in terms of the accelerations in the longi-
tudinal (x) and transversal (y) directions, in a total of 32 measure-
ment points, organized in two measurements setups. Each setup
involves the use of 4 reference sensors, placed on the deck (posi-
tions 4, 8 and 14) and on pier P7 (position 24), and 14 mobile sen-
sors, that changed their position from one setup to the other.
In the first measurement setup the mobile accelerometers were
positioned on the deck, on one of the side guards, in transversal
direction. Otherwise, in the second setup, the accelerometers were
installed at half-height of the piers and positioned in longitudinal
direction. Fig. 13 shows the location of the reference accelerome-
ters and the position of the mobile accelerometers for measure-
ment setups 1 and 2.
Data acquisition was performed using a cDAQ-9172 system
from NI (National Instruments) equipped with 24-bit resolution
IEPE type analogue input modules (NI 9234). The time series have
been acquired during periods of 10 min, with a sampling frequency
of 2048 Hz, and subsequently decimated to 256 Hz frequency. The
sampling frequency equal to 2048 Hz corresponds to the lowest
sampling rate allowed by NI 9234 module, which, posteriorly,
enabled a proper decimation of the signal. The connection of
accelerometers to the side guards and piers was carried out by
means of steel plates or metallic angles glued or bolted to the stone
surface. Fig. 14a and b shows the details of accelerometers located
on the side guard of the deck and on the front of one of the bridge’s
piers, respectively. The accelerometers placed on the piers were
installed with the help of a platform lift with extendable arm.
The typical extreme values of the deck’s accelerations, in
transversal direction, are approximately equal to ±0.25 mg, and
at the piers, in longitudinal direction, are approximately equal to
±0.12 mg.
3.2. Modal parameter identification
The identification of modal parameters of the bridge was per-
formed by the application of the Enhanced Frequency Domain
Decomposition method (EFDD) available at ARTeMIS software [36].
The EFDD method allows identifying the mode shapes by locat-
ing the peaks of the singular values decomposition plots calculated
from the spectral density functions of the measured responses
[37].
In EFDD method, the Single Degree-of-Freedom (SDOF) Power
Spectral Density function, identified around a peak of resonance,
is taken back to the time domain using the inverse discrete Fourier
transform. The natural frequency is obtained by determining the
number of zero-crossing as a function of time, and the damping
by the logarithmic decrement of the corresponding SDOF normal-
ized auto correlation function [37].
Each SDOF function is estimated using the shape determined by
peak picking and a reference singular vector is set for a further cor-
relation analysis based on the Modal Assurance Criterion (MAC).
Therefore, the MAC value between the reference singular vector
and a singular vector for each particular neighbourhood frequency
362 C. Costa et al. / Engineering Structures 123 (2016) 354–371

Mode 1 (f1T = 1.94 Hz) Mode 2 (f2T = 2.25 Hz)

Mode 3 (f3T = 2.58 Hz) Mode 4 (f1L = 2.89 Hz)

Mode 5 (f4T = 3.14 Hz) Mode 6 (f5T = 3.81 Hz)

Mode 7 (f6T = 4.57 Hz) Mode 8 (f2L = 5.12 Hz)

Mode 9 (f7T = 5.33 Hz)

Fig. 12. Bridge frequencies and modal shapes of numerically obtained vibration modes.

line, is computed. If the MAC value of this vector is above a user-
specified MAC rejection level, the corresponding singular value is
included in the description of the SDOF function. The lower this
MAC rejection level is, the larger the number of singular values
included in the identification of the SDOF function will be.
A more detailed explanation of EFDD method can be found in
reference [37].
Fig. 15 presents the curves of the average normalized singular
values of the spectral density matrices of two experimental config-
urations of the ambient vibration test that include the accelera-
tions measured in transverse (Fig. 15a) and longitudinal
(Fig. 15b) directions, and where the peaks corresponding to 9 of
the global vibration modes of the bridge can be identified.
Fig. 16, illustrates, in perspective, the identified modal configu-
rations with an indication of the average values of frequencies (f)
and damping coefficients (n). The analysis of mode shapes allows
identifying joint bending movements of the deck and piers in
transverse (1T to 7T) and longitudinal (1L and 2L) directions with
very good definition. Mode shapes 1L and 2L were defined at deck
level using an approximate linear extrapolation from the modal
coordinates obtained at half-height of the piers. The average values
of the damping coefficients are between 1.34% and 3.61%.
 C. Costa et al. / Engineering Structures 123 (2016) 354–371 363

Fig. 13. Measurement setups 1 and 2: location of the measurement points on the deck and piers.

Fig. 14. Details of the installation of accelerometers: (a) on the side guard of the deck, (b) on the front of the pier.

4. Calibration of the numerical model quencies of vibration and the corresponding mode shapes are
obtained. The realization of this last step in ANSYS environment
To minimize the differences between numerical and experi- constitutes an upgrade in relation to the original methodology pro-
mental modal responses, a calibration process was performed. This posed by Ribeiro et al. [38] and allowed a significant enhancement
process involved two stages: first, a sensitivity analysis, in order to on its computational efficiency.
identify the numerical parameters that more influence the modal In the Matlab software, based on the experimental and numer-
responses, and second, an optimization, which involved the appli- ical modal information, the mode pairing between numerical and
cation of an iterative method based on genetic algorithms, origi- experimental modes is performed using MAC parameter [41],
nally developed by Ribeiro et al. [38] and adapted to the case of and the values of the natural frequencies and the corresponding
Durrães bridge. MAC values are exported in text format.
Finally, the OptiSlang software, based on an objective function
and on the application of an optimization technique supported
4.1. Methodology by a genetic algorithm, estimates a new set of parameters focused
on the minimization of the objective function residuals. The gener-
The iterative calibration method based on a genetic algorithm ation of these new individuals is based on crossover, mutation and
involves the use of three software packages: ANSYS [16], Matlab replacement operations. This procedure is repeated iteratively
[39] and Optislang [40]. Fig. 17 shows the flowchart illustrating until the maximum number of generations is reached.
the computational implementation of the method with an indica- The objective function f comprises two terms, the first related to
tion of the operations performed by each software package. the natural frequencies and the second related to the MAC values:
In the ANSYS environment the FEM numerical model is defined  num 
X f exp
n modes
 fi nX
modes
  
based on a set of initial parameter values h1, h2, . . ., hk, where k is the f ¼a i
þb MAC /exp ; /num  1 ð1Þ
exp i i
number of individuals in each generation. The selection of calibra- i¼1
fi i¼1
tion parameters is based on the results of a previous sensitivity
exp num
analysis. The sets of parameter values of generation 1 are randomly where f i and f i are the experimental and numerical frequencies
generated by applying the Latin Hypercube Sampling method. for mode i, /exp
i and /num
i are the vectors containing the experimen-
After the generation of each model and based on their numerical tal and numerical modal information regarding mode i, and a and b
parameters a modal analysis is performed, from which the fre- are weighing factors of the terms of the objective function.
364
Fig. 15. Average normalized singular values of the spectral density matrices obtained from the experimental setup with mobile sensors located in the (a) transverse and (b)
longitudinal directions.
4.2. Sensitivity analysis
The sensitivity analysis intent was the selection of parameters
that most influence the values of frequencies and MAC, and there-
fore should be considered in the subsequent phase of optimization.
In this study a global sensitivity analysis was used, where all
parameters vary simultaneously, allowing to obtain, in a single
attempt, the sensitivities between the parameters and responses
from a set of samples generated by Latin Hypercube Sampling
method [42].
In Latin Hypercube Sampling method the domain associated to
each parameter is divided into intervals with equal probability of
occurrence. Each interval is defined by the value of the parameter
corresponding to its centre of gravity and in accordance with the
correspondent probability density function, which, in this case,
were all assumed as uniform. The choice of each of the intervals
is performed randomly, so each interval is counted only once for
sampling generation [43].
The sensitivity analysis results are based on Spearman correla-
tion coefficients (rSxy) that express the correlation between two vec-
tors of samples, x and y, based on their rank order vectors, R(xi) and
R(yi), by the expression [40]:
Pn  
i¼1 Rðxi Þ  RðxÞ Rðyi Þ  RðyÞ
r Sxy ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn  2 Pn 
i¼1 Rðxi Þ  RðxÞ i¼1 Rðyi Þ  RðyÞ
C. Costa et al. / Engineering Structures 123 (2016) 354–371

where RðxÞ 
and RðyÞ are the average values of the rank order vectors
R(xi) and R(yi), respectively, with i = 1, 2, . . ., n, where n is the number
of samples of the variables.
The sensitivity study involved three tests, performed sequen-
tially, each one associated with different approaches regarding
the number and types of numeric parameters considered.
All analyses were performed using a stochastic sampling tech-
nique based on 1000 samples generated by the Latin Hypercube
Sampling method and samples with MAC values below 0.50 were
skipped.
4.2.1. Approach 1
The first calibration approach involved considering eight of the
numerical model parameters, namely: the elastic modulus of the
masonry superstructure elements (Em), the corresponding lower
and upper bounds for the backfill (Ebl and Ebu) and the elastic mod-
ulus of the embankment (Ee); the densities of the masonry (dm),
the corresponding lower and upper for backfill (dbl and dbu) and
that of the embankment (de). The ranges for each parameter were
considered according to the lower and upper limits given in
Table 2.
 Fig. 18 shows the results of the Approach 1 sensitivity analysis
through a matrix of linear Spearman correlation coefficients. The
2ffi ð2Þ
correlation coefficients situated in the range [0.30; 0.30] were
excluded from graphical representation.
 C. Costa et al. / Engineering Structures 123 (2016) 354–371 365

Mode 1T f = 1.85 Hz | ξ = 2.83 % Mode 2T f = 2.08 Hz | ξ = 2.43 %

Mode 3T f = 2.41 Hz | ξ = 2.40 % Mode 1L f = 2.50 Hz | ξ = 3.61 %

Mode 5T f = 2.79 Hz | ξ = 3.37 % Mode 6T f = 3.31 Hz | ξ = 2.51 %

Mode 7T f = 3.83 Hz | ξ = 1.81 % Mode 2L f = 4.11 Hz | ξ = 2.36 %

Mode 8T f = 4.33 Hz | ξ = 1.34 %

Fig. 16. Experimentally measured modal parameters and vibration mode shapes.

The correlation matrix shows that the masonry elastic modulus
of superstructure elements, including arches, spandrel walls, piers
and abutments is the most sensitive parameter to responses, espe-
cially to vibration frequencies, with most values of the correlation
coefficients near 1.0. On its turn the embankment elastic modulus
also has a significant influence on most of the MAC parameter val-
ues. The high values of correlation coefficients associated with
these parameters, especially the masonry elastic modulus, foresee
the possibility of a stable estimated value at a later optimization
stage, but it can render difficult estimating the values of the less
sensitive elastic parameters. The use of a single elastic parameter
for all masonry elements is also clearly insufficient to adequately
reproduce the different anomalies observed in visual inspection.
The parameters associated with the density of the bridge con-
stituent materials have a reduced sensitivity to all responses and
therefore may be excluded from the optimization phase.
4.2.2. Approach 2
The second approach was performed by replacing the elastic
modulus of masonry by specific elasticity parameters of each
structural element of the bridge. Thus four new numerical param-
eters were created representing the elastic modulus of the abut-
ments (Eam), the piers (Ep), the spandrel walls (Es) and the arches
(Ea). The ranges of these new parameters are in correspondence
366 C. Costa et al. / Engineering Structures 123 (2016) 354–371

Fig. 17. Flowchart of the iterative calibration methodology based on genetic algorithm.

Fig. 18. Matrix of linear Spearman correlation coefficients obtained from Approach
1. Fig. 19. Matrix of linear Spearman correlation coefficients obtained from Approach 2.
 C. Costa et al. / Engineering Structures 123 (2016) 354–371 367

with the lower and upper limits of the masonry elastic modulus – The embankment elastic parameter had a significant reduction
shown in Table 2. With this approach it is intended to take advan- of the values of the correlation coefficients with the responses
tage of the high sensitivity of the masonry elastic modulus, so that in terms of modal parameters. This reduction is related to the
consistent estimates are achieved for that parameter values for the fact that some of the new masonry parameters are now able
various bridge structural elements. Fig. 19 shows the results of the to locally change the shape of some vibration modes contrary
Approach 2 sensitivity analysis through a matrix of linear Spear- to what happened in Approach 1.
man correlation coefficients. From this analysis, the parameters
associated with the density of the constituent materials of the
4.2.3. Approach 3
bridge have been excluded.
In the third approach new numerical parameters were defined
The analysis of the results allows the following conclusions:
which allowed improving some of the limitations identified in
Approach 2, namely:
– The elastic parameters of the piers and spandrel walls similarly
contribute to the modal responses, in particular for the majority
– The predictable inability to estimate the elastic parameters
of the transverse frequencies of the bridge, and therefore in the
associated with each structural element of masonry and back-
optimization phase there can be different combinations of these
fills, resulting in the need to aggregate back its elastic
parameters that lead to the same solution in terms of optimiza-
properties.
tion of the problem.
– Incorporate the variability of the materials’ parameters over the
– The elastic parameters of the arches and abutments contribute
bridge development, reflecting the anomalies observed in the
less significantly to the modal responses of the bridge in the
visual inspection described in Section 2.2.
transverse direction. Regarding arches, given their location
and layout, the influence of their elastic parameters should be
Thus, eight new parameters have been defined (K1 to K8) repre-
more relevant to the modal parameters in the vertical direction.
senting the correction factors of the elastic modulus of the ele-
In relation to the abutments, and given their location at the
ments in masonry, including the arches, spandrel walls, piers and
ends of the bridge, the influence of their elastic parameters is
abutments, located in different areas of the bridge. For this pur-
restricted mainly to the mode shapes with higher amplitudes
pose, eight zones were defined corresponding to the grouping of
in these areas.
– The elastic parameters of both backfills present low values of
the correlation coefficients with the responses. For the lower
backfill its inability to influence the modal parameters of the
bridge in the transverse direction is due: (i) to its location inside
the pier section which is unfavourable to control the transverse
stiffness of the masonry-filled composite section, and (ii) to
their elastic properties being of a much lower order of magni-
tude to those of the masonry.

Fig. 20. Matrix of linear Spearman correlation coefficients obtained from Approach
3. Fig. 21. Values of the numerical parameters for the optimization runs GA1 to GA4.
368 C. Costa et al. / Engineering Structures 123 (2016) 354–371
masonry elements belonging to each two consecutive spans. It was
also defined a new parameter (Kf) representing the correction fac-
tor of the elastic modulus of the upper and lower backfill through-
out the length of the bridge.
The K1 to K8 and Kf parameters have a variation range between
0.40 and 1.40, and the correction factors with values equal to 1.00
correspond to the situation wherein the elastic parameters assume
initial values listed in Table 2.
Fig. 20 shows the results of the Approach 3 sensitivity analysis
through the corresponding matrix of linear Spearman correlation
coefficients.
From the analysis of the correlation matrix it can be concluded
that, except for the correction factor of the backfill elastic modulus,
all other parameters show to be sensitive to the responses with
similar values of correlation coefficients between them, situated
in the interval 0.40–0.80, allowing, therefore, to predict a stable
optimization process.
4.3. Optimization
The optimization phase allowed obtaining the parameters val-
ues that minimize the differences between the numerical and
experimental modal responses, and involved the application of
an optimization technique based on a genetic algorithm as
described in [38].
The optimization of the bridge model involved the use of 8
design variables and 18 modal responses, 9 frequencies and 9
MAC parameters. The genetic algorithm was based on an initial
Fig. 22. Errors between the numerical frequencies, before and after updating, in relation to the average value of the experimental frequencies.
Fig. 23. Comparison of the MAC values, before and after the updating of the numerical model.
population consisting in 30 individuals and 150 generations, for a
total of 4500 individuals. The initial population was randomly gen-
erated by Latin Hypercube Sampling method. In this algorithm the
number of elites was equal to 1 and the number of substitute indi-
viduals was also defined equal to 1. The crossing rate was consid-
ered equal to 50% and the mutation rate was set equal to 15% with
a standard deviation, variable along the optimization, between
0.10 and 0.01. The objective function weights a and b were consid-
ered equal to 1.10 and 0.90, respectively.
The optimal values of the parameters were obtained based on
the results of four independent optimization runs (GA1 to GA4)
with different initial populations.
Fig. 21a and b represents plots of ratios between values of
numerical parameters K1 to K8 and the corresponding limits (indi-
cated in Section 4.2.3) for the optimization runs GA1 to GA4. A
ratio of 0% means that the parameter coincides with the lower
limit. A ratio of 100% means that it coincides with the upper limit.
The obtained values for the numerical parameters are also indi-
cated in brackets next to the corresponding plots.
Optimization results show that, in most areas, there is a trend of
reducing the structural elements stiffness. This finding is in line
with the results of visual inspection in which several anomalies
have been identified throughout the bridge development, as shown
in Fig. 10. The bridge areas where there is a greater tendency of
stiffness degradation are associated with the parameters K3, K5
and K8, which correspond to the spans 5–6, 9–10 and 15–16,
respectively. Some of these areas include piers and arches where
more serious anomalies were identified (see Section 2.2) from
 C. Costa et al. / Engineering Structures 123 (2016) 354–371 369

the visual inspection, fact that contributes for consistency of the
herein obtained results.
The grouping of structural elements with anomalies together
with other elements where no significant anomalies were identi-
fied complicates the interpretation of results, especially the corre-
lation between the optimal parameters’ values and the visual
inspection results.
In general, the results from the different optimization cases are
very close together, with variations lower than 16%, demonstrating
the stability and robustness of the genetic algorithm.
Fig. 22 shows the error values between numerical and experi-
mental frequencies, before and after calibration, taking as refer-
ence the values of the experimental frequencies. The results after
calibration are referred to the optimization run GA3, which is asso-

Mode 1 (1.85 Hz | 1.73 Hz) Mode 2 (2.08 Hz | 1.98 Hz)

Mode 3 (2.41 Hz | 2.35 Hz) Mode 4 (2.50 Hz | 2.54 Hz)

Mode 5 (2.79 Hz | 2.79 Hz) Mode 6 (3.31 Hz | 3.38 Hz)

Mode 7 (3.83 Hz | 3.97 Hz) Mode 8 (4.11 Hz | 4.43 Hz)

Mode 9 (4.33 Hz | 4.63 Hz)

Fig. 24. Comparison between the experimental and numerical mode shapes, after updating and considering optimization run GA3.
370 C. Costa et al. / Engineering Structures 123 (2016) 354–371
ciated to the lowest residual of the objective function. The average
error of the frequencies of global modes decreased from 14.5%,
before calibration, to 4.0% after calibration.
Fig. 23 shows a comparison of the MAC values before and after
calibration of the numerical model. The average MAC value passed
from 0.901, before calibration, to 0.926 after calibration, and all
MAC values are higher than 0.890 after calibration.
Finally, Fig. 24 shows the comparison of experimental and
numerical mode shapes after calibration. In the figure there are also
indicated, in brackets, the values of the experimental and numerical
natural frequencies, respectively, for optimization case GA3.
5. Conclusions
This paper addressed several aspects about the development
and experimental calibration of a finite element method (FEM)
based numerical model of a stone masonry arch bridge existing
in normal service conditions in the Minho railway line connecting
Porto to the up-northwest of Portugal.
The experimental identification of modal parameters was based
on an ambient vibration test, which allowed estimating natural fre-
quencies and mode shapes of several vibration modes in transverse
and longitudinal directions resorting to a frequency domain iden-
tification technique.
The experimental results obtained from material mechanical
characterization tests made in situ and in laboratory was a quite
important and useful task to ensure a better estimate of the initial
values of the material mechanical parameters adopted in the FEM
model.
The sensitivity analysis revealed the importance of an adequate
strategy for defining the number and types of calibration parame-
ters especially in situations with limited experimental information
and where several parameters similarly contribute to the modal
responses, therefore rendering more difficult the estimate of stable
optimal solutions.
The model calibration results evidenced quite good agreement
between numerical response and experimental measurements. In
fact, comparing with the results of a numerical model defined prior
to the calibration, the calibrated model has shown significant
improvements on simulating modal parameters, as proven by
smaller deviations of numerically obtained frequencies relative to
experimental ones and by increased MAC coefficients of numerical
and experimental modal configurations. The optimization results
show a tendency of reducing the stiffness of the structural ele-
ments of the bridge which is in accordance with the results of
visual inspection in which several anomalies have been identified
throughout the bridge development.
In future studies, the calibrated numerical model of the bridge
will be used to assess the dynamic behaviour of the train-track
coupled system, particularly for freight trains, in order to draw
conclusions regarding the performance of the bridge in terms of
structural safety and track safety, concerning the track and
wheel-track contact stability. These dynamic studies are based
on advanced train-track-structure interaction models that also
include a calibrated model of the freight train and track irregular-
ities measured by a track inspection vehicle. Additionally it is
planned to perform a dynamic test under traffic actions in order
to validate the dynamic responses of the train-bridge system.
Acknowledgements
This work includes research carried out with the financial sup-
port of the Portuguese Science Foundation (FCT – Fundação para a
Ciência e Tecnologia) through the project PTDC/ECM-
EST_1691/2012 entitled ‘‘Experimental and Numerical Characteri-
zation of Structural Behaviour of Stone Masonry Arch Bridges
under Railway Traffic Loading - Application to Existing Portuguese
Bridges (StonArcRail)”. The authors which to thank engineers Ana
Isabel Silva, Hugo Patrício and Nuno Lopes from REFER for their
collaboration and for the precious help on providing information
about the bridges. The authors’ gratitude is also expressed to Dr.
Nuno Pinto and Mr. Valdemar Luís, both technicians of the LESE
laboratory, for their indispensable assistance during the prepara-
tion and execution of experimental tests.
References
[1] Orbán Z. UIC project on assessment, inspection and maintenance of masonry
arch railway bridges. In: Proceedings of the 5th international conference on
arch bridges (ARCH 07), Madeira, Portugal, 2007.
[2] Olofsson I, Elfgren L, Bell B, Paulsson B, Niederleithinger E, Jensen J, Feltrin G,
Täljsten B, Cremona C, Kiviluoma R, Bien J. Assessment of European railway
bridges for future traffic demands and longer lives – EC project ‘‘Sustainable
Bridges”. Struct Infrastruct Eng 2007;1:93–100.
[3] McCann D, Forde M. Review of NDT methods in the assessment of concrete and
masonry structures. NDT and E Int 2001;34(2):71–84.
[4] Orbán Z, Gutermann M. Assessment of masonry arch railway bridges using
non-destructive in situ testing methods. Eng Struct 2009;31(10):2287–98.
[5] Roberts T, Hughes T. Serviceability of masonry arch railway bridges. In:
Proceedings of the 5th international conference on arch bridges (ARCH 07),
Madeira, Portugal, 2007.
[6] Brencich A, Sabia D. Experimental identification of a multi-span masonry
bridge: the Tanaro Bridge. Constr Build Mater 2008;22(10):2087–99.
[7] Brencich A, Sabia D. Structural identification of masonry arch bridges. In: Proc.
int. operational modal analysis conference 2009 (IOMAC 2009).
[8] Arêde A, Costa P, Costa A, Costa C, Noites L. Monitoring and testing of a new
stone masonry arch bridge in Vila Fria, Portugal. In: Proceedings of the 5th
international conference on arch bridges (ARCH 07), Madeira, Portugal, 2007.
[9] Costa C, Arêde A, Costa A, Caetano E, Cunha A, Magalhães F. Updating
numerical models of masonry arch bridges by operational modal analysis. Int J
Archit Heritage 2015. http://dx.doi.org/10.1080/15583058.2013.850557.
[10] Costa C, Arêde A, Costa A. Mechanical characterization of the constituent
materials of stone arch bridges. In: Proceedings of the 9th international
masonry conference (9IMC), Guimarães, Portugal, 2014.
[11] Srinivas V, Sasmal S, Ramanjaneyulu K, Ravisankar K. Performance evaluation
of a stone masonry-arch railway bridge under increased axle loads. J Perform
Constr Facil 2014;28:363–75.
[12] De La Hire P. Sur la construction des voûtes dans les édifices. Mémoires de
mathématique et de physique de l’Académie royale des sciences; 1712.
[13] Pippard AJS. The approximate estimation of safe loads on masonry bridges. Civ
Eng War 1948;1:365–72.
[14] Heyman J. The safety of masonry arches. Int J Mech Sci 1969;11(4):363–85.
[15] Fanning PJ, Boothby TE, Roberts BJ. Longitudinal and transverse effects in
masonry arch assessment. Constr Build Mater 2000;15(1):51–60.
[16] ANSYSÒ Academic Research, Release 15.0, Help System, ANSYS Inc; 2014.
[17] Frunzio G, Monaco M, Gesualdo A. 3D FEM analysis of a roman arch bridge.
Hist Constr 2001:591–8.
[18] Cavicchi A, Gambarotta L. Collapse analysis of masonry bridges taking into
account arch-fill interaction. Eng Struct 2005;27(4):605–15.
[19] Arani K, Zandi A. Protection of an old stone masonry arch bridge against railway
impact. In: Protection of historical buildings (PROHITECH 09), Rome, Italy.
[20] Costa C, Arêde A, Costa A. Detailed FEM modelling of stone masonry arch
bridges under road traffic moving loads. In: Proceedings of the 3rd ECCOMAS
thematic conference on computational methods in structural dynamics and
earthquake engineering (COMPDYN), Corfu, Greece.
[21] CEA, Presentation et utilization de Cast3M by E. Le Fichoux, Commissariat à
l’Énergie Atomique; 2011. Available at: <http://www-cast3m.cea.fr>.
[22] Costa C, Arêde A, Morais M, Costa A. FEM and DEM detailed modelling of stone
masonry arch bridges for the assessment of load-carrying capacity. In:
Proceedings of the 1st international conference on structural integrity
(ICSI1). Funchal: ESIS; 2015. Procedia Eng 2015;114:854–861.
[23] Andersson A. Load capacity assessment and strengthening of a railway arch
bridge with backfill. In: Proceedings of the IABSE Spring Conference; 2013.
[24] SOLVIA Finite Element System. User manual. Vasteras, Sweden: SOLVIA; 2003.
[25] Domede N, Sellier A, Stablon T. Structural analysis of a multi-span railway
masonry bridge combining in situ observations, laboratory tests and damage
modelling. Eng Struct 2013;56:837–49.
[26] Friswell M, Mottershead J. Finite element model updating in structural
dynamics, vol. 38. Springer Science & Business Media; 1995.
[27] Bayraktar A, Altunisßik A, Birinci F, Sevim B, Türker T. Finite-element analysis
and vibration testing of a two-span masonry arch bridge. J Perform Constr Facil
2009;24(1):46–55.
[28] Bayraktar A, Birinci F, Altunısßık A, Türker T, Sevim B. Finite element model
updating of Senyuva historical arch bridge using ambient vibration tests. Baltic
J Road Bridge Eng 2009;4(4):177–85.
[29] Arêde A, Costa C, Morais M, Topa Gomes A, Silva R. Avaliação experimental do
comportamento dos componentes e dos materiais de uma ponte ferroviária
 C. Costa et al. / Engineering Structures 123 (2016) 354–371
em alvenaria de pedra. In: Proceedings of the 4° Congresso Nacional -
Segurança e Conservação de Pontes (ASCP2015). Lisboa: LNEC; 2015.
[30] Binda L, Tiraboschi C. Flat-Jack Test: a slightly destructive technique for the
diagnosis of brick and stone masonry structures. Int J Restor Build Monuments
1999:449–72.
[31] ASTM1197-91. In situ measurement of masonry deformability properties using
flat-jack method. ASTM Standard C 1197-91; 1991.
[32] RILEM-LUM.D.2/3. In situ stress and strength/elasticity tests on masonry based
on the flat jack; 1990.
[33] NFP94-110-1. Ménard Sols: reconnaissance et essais, Essai presiométrique
Ménard; 2000.
[34] Briaud J. The Pressuremeter. Trans Tech Publications Roterdão, ed. Balkema;
1992.
[35] Costa C, Ribeiro D, Jorge P, Silva R, Arêde A, Calçada R. Avaliação Experimental
e Numérica dos Parâmetro Modais da Ponte Ferroviária de Durrães. In:
Proceedings of the 5as Jornadas Portuguesas de Engenharia de Estruturas (JPEE
2014), Lisbon, Portugal, 2014.
371
[36] ARTeMIS Extractor Pro – Academic License. User’s manual. Aalborg,
Denmark: ARTeMIS; 2009.
[37] Brincker R, Zhang L, Andersen P. Modal identification from ambient responses
using frequency domain decomposition. In: Proceedings of the IMAC XVIII, San
Antonio, USA, 2000.
[38] Ribeiro D, Calçada R, Delgado R, Brehm M, Zabel V. Finite element model
updating of a bowstring-arch railway bridge based on experimental modal
parameters. Eng Struct 2012;40:413–35.
[39] MATLAB. Mathworks. USA: Getting Started Guide Natick; 2011.
[40] OptiSLang. OptiSLang – the optimizing structural language. 3.2 ed. Weimar,
Germany; 2008.
[41] Allemang RJ. The modal assurance criterion – twenty years of use and abuse. J
Sound Vib 2003;37:14–21.
[42] Saltelli A, Tarantola S, Campolongo F, Ratto M. Sensitivity analysis in practice:
a guide to assessing scientific models. John Wileys & Sons Ltd; 2004.
[43] Stein Michael. Large sample properties of simulations using Latin hypercube
sampling. Technometrics 1987;29(2):143–51.