You are on page 1of 13

Advances in Engineering Software 101 (2016) 136148

Contents lists available at ScienceDirect

Advances in Engineering Software


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

Modelling and strength evaluation of masonry bridges using terrestrial


photogrammetry and nite elements
M.E. Stavroulaki a,, B. Riveiro b, G.A. Drosopoulos c, M. Solla b, P. Koutsianitis c,
G.E. Stavroulakis c
a
b
c

Technical University of Crete, School of Architecture, Applied Mechanics Laboratory, GR-73100 Chania, Greece
University of Vigo, School of Industrial Engineering, Department of Engineering Materials, Applied Mechanics and Construction ES-36208 Vigo, Spain
Technical University of Crete, School of Production Engineering and Management, GR-73100 Chania, Greece

a r t i c l e

i n f o

Article history:
Available online 13 January 2016
Keywords:
Nondestructive evaluation
Photogrammetry
Finite element analysis
Damage mechanics
Ground penetrating radar
3D modelling

a b s t r a c t
Several numerical models are presented in this article, for the study of the ultimate behaviour of a real
stone arch bridge. For the exact representation of the geometry an integral and comprehensive survey
involving Terrestrial Photogrammetry and Ground Penetrating Radar is in order to provide a realistic 3D
geometric model for the subsequent mechanical analysis of the bridge. The accuracy of the photogrammetric method permitted detecting cracks in different areas and the GPR completed the geometric model
with information of hidden parts such as backll, arch ring thickness, etc. Finite element analysis models,
incorporating damage, elastoplasticity and contact, are then developed. Comparison between these models is considered in a single arch of the structure. The classical four hinges mechanism appears in the
arch. A model of the whole structure, where the arch and the ll are taken into account, is nally developed. Results show how damage is developed in the body of the arch, for loadings that include forces, or
vertical and transverse displacements in the supports.
2016 Civil-Comp Ltd. and Elsevier Ltd. All rights reserved.

1. Introduction
New photogrammetry techniques allow us measure structures
of complex shapes and create accurate models for further structural analysis. Within this paper we report on an application of
this technique on a model structure, the Cernadela Bridge in Spain.
The purpose of this paper is twofold. First practical diculties arising during the mentioned operation will be briey discussed and
practical structural analysis and evaluation tasks related to a masonry bridge will be presented. Furthermore open questions and
the needs for further development of the involved techniques will
be identied and listed.
Some rst related results have already been published in [1].
In the present article, new information regarding terrestrial photogrammetry is given and details for obtaining the geometry
of the ll over the arch are presented. That leads to an improved and more accurate model, including information from the

Corresponding author. Tel.: +30 2821028364.


E-mail addresses: mstavr@mred.tuc.gr, mestavr@gmail.com (M.E. Stavroulaki),
belenriveiro@uvigo.es (B. Riveiro), gdrosopoulos@isc.tuc.gr (G.A. Drosopoulos),
merchisolla@uvigo.es (M. Solla), panoskout@gmail.com (P. Koutsianitis),
gestavr@dpem.tuc.gr (G.E. Stavroulakis).
http://dx.doi.org/10.1016/j.advengsoft.2015.12.007
0965-9978/ 2016 Civil-Comp Ltd. and Elsevier Ltd. All rights reserved.

interior of the structure, which stands for the simulation of the


whole structure.
A signicant number of masonry arch bridges in Europe still
survive and some of them are still being used, therefore a detailed analysis of these monuments is of great interest. Masonry
arches consist of stone blocks and the mortar joints. Blocks have
high strength in compression and low strength in tension while
mortar has generally low strength. Other mechanical properties
(like Youngs modulus) are also different between the constitutive materials of these structures. Consequently, a great number of theories have been developed in the past, in order to
capture this variation in the mechanical properties of masonry
arches.
In this framework continuums as well as discontinuous (discrete) models are used and nite element analysis with commercial packages or limit analysis schemes are adopted. The distinction between continuum and discrete models is given later in this
study.
Sophisticated homogenisation, as well as multi-scale methods
have been proposed for the investigation of the behaviour of masonry structures. In [2] a method combining structural analysis and
homogenisation is proposed for the study of masonry. A third material which is a mixture of the bricks and mortar is introduced
and a multi-scale scheme is developed to investigate failure of the

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

masonry interfaces. A mortar constitutive law, which takes into


account the coupling of the damage and friction phenomena occurring during the loading history, is proposed in [3]. An effective
non-linear homogenisation procedure, based on the transformation
eld analysis and the nite element method, is then proposed. In
[4] and [5] a computational homogenisation approach is used to
depict localisation phenomena in masonry structures.
A lot of efforts towards establishing numerical methods for the
investigation of masonry structures in a structural level, have been
also appeared. A simple and quick procedure for the assessment
of the seismic vulnerability of masonry compounds, based on the
calibration on the basis of numerical analyses performed at different urban scale levels, has been proposed in [6]. A three dimensional computational model, based on the Discrete Element
Method (DEM) was proposed in [7]. It was used to investigate the
effect of the angle of skew on the load carrying capacity of several single span stone masonry arches. In [8] the usage of nonlinear beam elements with bre cross-section has been proposed
for modelling the ultimate behaviour of multi-span masonry arch
bridges. The interaction among the spans and the non-linear material behaviour can be described with low computational effort.
Vibration measurements have been used in [9] to investigate damage in masonry structures at an early stage. For this purpose, an
approach based on dynamic damage identication methods, has
been proposed. In [10] the static structural behaviour and the dynamic properties of an old masonry church have been investigated
using non-linear nite element analysis with proper constitutive
assumptions. The seismic vulnerability has been also evaluated using a pushover method.
From another point of view, limit analysis tools in the framework of constraint optimisation have been proposed. In [11] the
collapse load of discrete rigid block systems with frictional contact interfaces was computed, as a special constrained optimisation
problem (the so-called mathematical problem under equilibrium
constraints, MPEC). In [12] the limit analysis problem was formulated as an optimisation problem and a solution which involved
the use of a genetic algorithm was suggested. A discontinuous upper bound limit analysis approach with sequential linear programming and mesh adaptation has been proposed in [13] for the investigation of the behaviour of masonry double curvature structures.
The resulting combination of techniques allows us evaluate the
structural health of an existing complex masonry structure by
taking into account geometric data of high precision, mechanical
models of adequate complexity and elements of inverse analysis.
Further information, like material data from the interior of the
structure, is taken into account.
Consequently, the goal of this article is rst to point out the
interface between photogrammetry and structural analysis. To obtain a more realistic output of the structural behaviour, information taken from photogrammetry regarding damage of the arch
(cracks) is taken into account. Second, several continuum damage
and discrete models are used, in order to offer the analyst a holistic
insight concerning the tools which may be used for the structural
evaluation of masonry arches.
2. 3D modelling of Cernadela Bridge
In this article, a structural evaluation of a masonry bridge
in Spain, the Cernadela Bridge is considered. The creation of a
realistic geometric model involved several complementary nondestructive surveys. Overall, the methodology to create the geometric model to be used in the subsequent structural analysis comprises the following data:
- 3D model of the bridge; this model was created using photogrammetry due to the necessity of obtaining the external 3D

137

geometry the bridge. This was achieved by restitution of the


contours of each individual ashlar. This model has a double purpose: to provide information of the external envelope of the
bridge to serve as a basis for the FEM mesh and to visually
identify cracks in the masonry walls. Cracks can be visually detected when the joint between two ashlars is wider than the
surrounding joints.
- Cracks model; from the previous 3D model, those joints perpendicular to masonry courses, identied as cracks were saved
as a different point cloud where points were delineated using
polylines.
- Complementary geometric data of inner parts of the bridge provided by a Ground Penetrating Radar prospection. By georreferencing the GPR data to the coordinate system dened for the
photogrammetric survey, it was possible to add information of
the nonvisible elements of the bridge such as average thickness
of spandrel walls, thickness of arch ring and pavement, composition of backll, etc.
According to these models the methodology for the creation of
a realistic geometric model of Cernadela Bridge was divided in two
phases:
- Integral eld survey collecting both, photogrammetric GPR data
and subsequent data processing to create the three models related above.
- Creation of an accurate and detailed geometric model integrating cracks and the characteristics provided by GPR.
This section aims to summarise the different surveys conducted
in the Cernadela bridge: photogrammetric and GPR surveys. Finally
the methodology that permits integrating all the geometric models
in a single FEM mesh is presented.
2.1. Field works: photogrammetry and ground penetrating radar
surveys
Terrestrial photogrammetry has been reported as a good candidate for the performance of accurate geometric surveys of historical structures due to high level of accuracy and detail provided
by the technique. Close range photogrammetry was deeply studied and evolved during last decade in many different elds and
particularly in architecture and civil engineering [14, 15], but its
application to the study of historical structures is not a question
solved a priori. Historical elements have singularities (complex and
large geometry, they are heritage elements that require nondestructive evaluation, etc.), which demand the development of specic methodologies in order to achieve its complete metric survey.
To complete the information provided by photogrammetry,
which exclusively refers to visible information, ground penetrating radar (GPR) is proposed. GPR is a geophysical method that has
been established as one of the most recommended non-destructive
methods for routine sub-surface inspections. The technique has
proven its suitability for providing high image quality results of the
interior of the structures, and from the mid-1970s there have been
published numerous studies applied to many aspects related to
civil engineering eld. Regarding the evaluation of masonry structures, the GPR technology has demonstrated its potential to document and measure different inner structural characteristics, such
as the dimensioning of wall thicknesses, the detection of internal
faults like voids and cracks, as well as pathologies in construction,
and also to locate hidden structures and former geometries [16,
17].
2.1.1. Photogrammetric survey and data processing
Photogrammetry is dened as a method that allows the geometry of objects to be reconstructed from images, where the object

138

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

has previously impressed. This is possible through the establishment of geometrical relationships between objects coordinates (in
3D space) and image coordinates (in 2D space) into a perspective system, governed by the collinearity condition that establishes
that, at the time of exposure, a point in the object space, the perspective centre and the image coordinates of the point all lie in
common straight line [18].
In essence, two main steps are involved in the solution of that
process. First, the parameters that dene the perspective system
that produces the impression of images are required, which are
achievable through the camera calibration. Then, relative orientation consists of the determination of the relative position and orientation of those perspective systems (cameras) involved in the
photogrammetric model, to be nally transformed to the global coordinate system once scale factor of the model is known.
Photogrammetric data acquisition: an SLR digital camera manufactured by Canon, model EOS 10D, mounted with a 20 mm
Nikkor lens was used for image recording. Camera calibration values, which are very important for a reliable application, can be
seen in [18]. Another factor is related to the PSP, and consequently,
to the distance between camera and bridges surface. During the
image acquisition the surveyors tried to maintain operational distance with values around 15 m. According to camera parameters
the expected PSP would have a value of 5 mm.
Working under the operational distances mentioned, the eld
of view (FOV) of the camera does not cover the whole structure.
Because of this reason the survey of the structure had to be split
in smaller models, being each model composed of three images
taken from different points of view.
The photogrammetric survey of Cernadela Bridge was based on
the principles of convergent photogrammetry where images maintain optimal convergence angles between main directions of cameras of 90. In this sense, 32 individual models were required to
complete the 3D modelling of the whole structures envelope.
Articial targets were distributed in all around the bridge surface so the coordinates of the minimum number of control points
in object space could be collected by topographic equipment. A
total of 100 control points were measured by means of a Total Station Leica TCR1102. The topographic measurements were
subsequently introduced into the photogrammetric workstation to
achieve the absolute orientation. This procedure guarantee not only
to have the 3D model scaled and levelled, but also to control accuracy of nal 3D model by comparing with the truth data provided
by total station.
Also, the position of target marks for the GPR survey where
identied manually in the photogrammetric model to guarantee
the registration of the internal proles on the model envelope
(bridge surface).
Photogrammetric data processing: the images collected in the
eld and the coordinates of control points measured by total station were downloaded to the photogrammetric workstation Photomodeler Pro. The route followed in this application was the
following: 1) Import of images corresponding to each independent model. Each model was composed of 3 or 4 convergent images with optimal convergent angles of 90. 2) Inner orientation.
The geometry of the perspective system used for the impression
of images was reconstructed through the information of camera
calibration. 3) External orientation. The relative orientation was
performed for each independent model by identifying 6 common
points in the homologous images that dened each model. The information of these points in image space allowed solving the relative position between cameras and also those points in the model
space (neither scale nor absolute orientation yet). The 32 3D models orientated in this step belong to different coordinate systems,
and consequently all need to be joined together in order to create
the 3D model of the whole structure. This was done by marking

and transferring common points (at least 3) in the overlapping areas between adjacent models. After this, a relative model of the
whole structure was formed; 4) Absolute orientation. After referring corresponding control points in both topographic data and relative model data, the 3D model of Cernadela Bridge in the global
coordinate system was completed. 5) Restitution. Once all the images were externally oriented the next step consisted of restitution
of all those points that represent the geometry of the elements required for the subsequent structural analysis of the bridge.
2.1.2. Complementary geometric data provided by GPR
As presented in the previous paragraphs, GPR was used as a
complementary source of data to have information of those nonvisible areas of the bridge. Particularly, information of zonication of backll, pavement and ring stone thickness, etc. was estimated. The trajectory of the GPR antennas was accurately marked
by means of articial targets in order to register the path followed
by the GPR into the same coordinate systems used for the photogrammetric model. This operation was performed at the pathway of the bridge. Additionally, different grids were marked also
using articial targets into one of the vaults to provide a more accurate measure of arch thickness and so have an estimation of the
error of estimating arch thickness from the survey made from the
path, and nally, grids were also marked into the pillars where the
conguration 4 of GPR, that is presented in the next sections, was
applied.
GPR survey: the GPR data were collected using a RAMAC/GPR
system from MALA Geoscience. Three different frequencies and acquisition parameters were selected depending on the application
and the data required. The conguration 1 in Table 1, using a central frequency of 250 MHz, was selected to map the lling in the
interior of the whole structure, as well as to determine the paving
thicknesses. Additionally, a 500 MHz antenna (conguration 2) was
chosen to improve imaging resolution. Thus, the paving level was
better recognised and more accurate measurements of thickness
could be obtained. The conguration 3 was carried out to provide
the ring stone thicknesses. This data acquisition was performed in
the longitudinal direction to the bridge structure, through the entire vault intrados surface; while both congurations 1 and 2 were
conducted along the pathway of the bridge in the longitudinal direction. Finally, the conguration 4, in which the GPR proles were
gathered in the vertical direction through the accessible wall piers
of the bridge, allowed for a proper assessment of the presence
of lling or solid granite (ashlar) inside the structure. To measure
the prole lengths, and for trace-interval distance calculation, an
odometer wheel (encoder) attached to the antenna was used.
GPR data processing: all of the collected proles were ltered
before interpretation to correct the down shifting of the signal
caused by air-ground interface and to amplify the received signal as well as to reduce clutter and unwanted noise in the raw
data (both low- and high-frequency noise in the temporal and spatial directions). The objective was to enhance the extraction of
information from the received signals and to produce a subsurface image that includes all of the features and/or targets of interest, which simplies the interpretation of the GPR data. Moreover,
topographic corrections provided by the photogrammetric model
were applied to the proles acquired through the vault intrados
surface. Migration ltering was also applied to the GPR data gathered through the wall piers in order to mitigate the diffraction hyperbolae produced by the heterogeneous lling and irregular ashlar, as well as joints between ashlars, which allowed for a better
recognition of layering and more appropriate estimation of depths
(or thicknesses). The data were processed with the ReexW software [19].
Thicknesses values were determined from Eq. (1). This value
is coincident with the distance travelled by the wave (d), and it

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

139

Table 1
Survey parameters assumed for GPR data acquisition by considering four different congurations based
on the data results desired.
Conguration

Frequency (MHz)

Trace-interval (cm)

Time windows (ns)

Samples/trace

1
2
3
4

250
500
800
500

5
2
1
3

220
100
55
100

566
677
554
710

Fig. 1. (a) 3D wireframe model of the whole structure of Cernadela Bridge. (b) Detailed model of second vault of the bridge with camera position and intersection rays of
some points. (c) Cracks measured in in the spandrel walls between arches 4 and 5. (d) 3D solid model created from the integration of GPR and cracks in the photogrammetryderived model.

is obtained by knowing the radar-wave velocity of propagation in


medium (v), and the travel-time distance (twt) to and from the
layer. For calculations, average radar-wave velocities of 11.0 and
12.0 cm ns1 were assumed for lling materials and granitic ashlar, respectively [17,20].

d=v

twt
2

(1)

2.2. Creation of geometric models: 3D solid model and cracks model


The previous non-destructive surveys provided the necessary
information to create the geometric basis for the FEM analysis.
Once that information is available, the next step consisted in its
optimal integration and exploitation. In that sense, the methodology for the creation of the geometric models involved: rst, to
have a basic 3D model of the bridge envelope; second, to add information of non-visible parts such as backll characterisation, arch
ring thickness along the vault, etc.; and nally, to add the cracks
identied during the photogrammetric restitution of ashlars to the
integral model of the bridge.

2.2.1. Basic 3D model


After the restitution of points dening the contours of all the
masonry blocks, the 3D model of Cernadela Bridge is composed of
more than 25 thousands points with averaged precisions for XYZ
components of (0.0 07; 0.0 08; 0.0 05) giving a RMSE of 12 mm. This
data comprises the 3D point cloud of the bridge. Fig. 1a shows the
3D wireframe model of Cernadela Bridge where all the ashlars contours. Fig. 1b represents camera conguration for the reconstruction of second vault of the bridge with intersection rays from different cameras.
Characterisation of inner parts of the 3D model: before the creation of the FEM mesh, the information extracted from the GPR
survey was used to update the geometric model. After having the
GPR proles geometrically aligned to the coordinate system of the
general 3D model, it was possible to complete the general model
with the information interpreted in the GPR data.
In general, the results obtained from the GPR data allowed obtaining inner geometrical characterisation that was previously unknown, namely: ring stone and paving thicknesses proles, as well
as the zonication of lling and solid granitic ashlar in the interior

140

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

Fig. 2. Geometrical data obtained from GPR. a) 250 MHz data showing the zonication of lling in the interior of the bridge, and paving thicknesses prole. b) 500 MHz
data interpreting the pavingll interface in detail, and the arch ring corresponding to the rst arch from the left side upstream. c) 800 MHz data obtained through the
intrados surface of the vault, which allowed identifying the ring stone thicknesses prole with more accuracy.

of the bridge (Fig. 2). The 250 MHz data collected along the pathway of the bridge was useful to obtain (Eq. (1)) the paving thicknesses prole (Fig. 2a). This interpretation was validated with the
detailed results produced by the 500 MHz data achieved through
pathway (Fig. 2b). Although the ring stone thicknesses can be appreciate from the 250 MHz data, and more pronounced from the
500 MHz data, the radargram generated when acquiring data with
the 800 MHz antenna through the internal intrados of the vault
provided the ring stone prole with more resolution and, subsequently, more accurate thicknesses values (Fig. 2c).
In terms of inner constructive materials, the 250 MHz data
recorded through pathway shown those areas containing more heterogeneous lling in the interior of the structure (Fig. 2a). Additionally, a 500 MHz GPR survey was carried out along the surface of the accessible wall piers, which demonstrated that such
structural components are composed by solid ashlar until approximately 1 m height (Fig. 3).
Creation of the 3D solid model: in order to integrate this
model into the procedures of structural analysis, the initial point
cloud and the information provided by the GPR have to be postprocessed together in order to create a 3D solid model before creating the FEM mesh. This process performed in two steps using
Geomagic Spark software: rst, the point cloud was converted into
a surface model through a triangulation process, and later, it was
converted into a solid model. The next step involved the partition
of the global 3D model with according to the characterisation provided by GPR, and so, vaults and backll were isolated before creating the FEM mesh.
Introduction of crack information: due to the accuracy of the
photogrammetric model, cracks were visually identied during the
restitution of ashlars. The delineation of cracks was exported as
an independently layer to Geomagic Spark, where it was used as
the pattern to create the partitions of the 3D solid model. Since,

there is only evidence of cracks in the intrados of some vaults,


the cracks were used to cut the vaults longitudinally according to
the cracks pattern. Evidence of cracking was also measured in one
of the spandrel walls (Fig. 1c), however this cracks are caused by
an out of plane deformation, which provoked a general opening
of masonry pieces. Even these cracks were documented, they were
not used to cut the 3D model because out of plane deformations
were not included in the further structural analysis considered in
this work.
Finally, the integral geometric model created was exported to
STEP format (Standard for the Exchange of Product Data) to be imported in the FEM software, Marc and Abaqus, for the work presented in this article.
3. Structural nite element analysis
3.1. Creation of cad model and nite elements
The geometry models initially developed from photogrammetry are mainly consisted of point clouds and lines. For this reason,
they are imported into appropriate computer aided design software with specialised surface processing tools, which are used for
the creation of complex surfaces, solids and sets of parts. This improved version of the geometry model is nally imported into classical nite element analysis packages for the structural assessment
of the bridge. Marc and Abaqus have been used in this article,
respectively.
3.2. Finite element analysis: interface modelling versus continuous
damage mechanics
The computational models which have been developed in the
past can be roughly divided into two large categories: (a) discrete
models and (b) continuum models.

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

141

Fig. 3. 800 MHz GPR data gathered through the wall pier among the two rst arches at the left margin from upstream. The layering of solid ashlar in depth is interpreted,
as well as the interface between ashlar and lling in height.

In a discrete model formulation the structure is divided into


large discrete deformable parts connected with interfaces. The behaviour of the contact surface in each interface is described by a
unilateral law, possibly with friction, while the discrete elements
are assumed to behave elastically. Detailed discontinuous nite element model incorporating principles taken from non-smooth mechanics like unilateral contact and friction, have been presented
among others in [18,2123].
From another point of view, the mechanical behaviour of continuum models is described by a nonlinear constitutive law, where
either the masonry is assumed to consist of a single material and
its behaviour is described by an inelastic theory (for instance an
appropriately modied damage model) [24], or the different mechanical behaviour between stone and mortar and the anisotropy
induced by them are taken into account on the basis of a homogenisation theory [25].
Experience accumulated from the discontinuous modelling approach, indicated the fact that consideration of potential cracks as
macroscopic interfaces, although is a quite realistic method for the
representation of the mechanical behaviour, however requires the
handling of dicult numerical schemes, in case the method is expanded to large, complex structures with more complicated pattern of interfaces. Moreover, the computational cost towards this
effort would be signicant for a large scale structure, for instance
a three-dimensional model of a multi-ring stone arch bridge. For
these reasons, a rst attempt for a correlation between the discrete
macroscopic approach described above and a continuous damage
model, was made by the authors of this study in [26].
In the next sections of the article, both continuum damage
models and discrete models are used for the determination of the
ultimate behaviour of the arches of the Cernadela stone bridge. An
elastoplastic law is applied to the ll.
3.3. Ultimate behaviour and collapse prediction
When a stone arch bridge is close to failure, a small increase
of the loading which is applied to the arch causes a signicantly
increased vertical displacement of it. The structure then reaches
its ultimate strength and the analysis is terminated. Thus, in the
framework of nite element analysis as the load is increased, damage arises and expanding in the body of the arch, while the force
displacement diagram tends to become horizontal, indicating that
the structure is close to collapse. This point, which is characterised
with at least one zero eigenvalue of the tangential stiffness matrix,
is at the end of a path of stable mechanical equilibrium which in
fact has been obtained with a monotonic application of the loading. The numerical tools can proceed further, following branches
of unstable solutions in analogy to post-buckling effects, using for
example tools like arc-length techniques. A careful path-following,
incremental-iterative solution procedure is followed in this paper,

Fig. 4. Uniaxial stressstrain law of the smeared crack damage model.

Fig. 5. Failure surfaces of the smeared crack damage model.

which in comparison with available predictions from other methods gives us condence that prediction of collapse corresponds to
real collapse and not to numerical failure.
It is noted that in the framework of the present work, nonlinear, incremental nite element analysis is used to depict damage, among others, in a real and complex masonry arch bridge.
From another point of view, one of the most common procedures
which can be used to evaluate the collapse mechanism and the ultimate load is related to limit analysis (see for instance [1113]).
3.3.1. Smeared crack damage model for the arches
In this article two similar continuum models have been used for
the structural analysis of the arches of the structure. The rst one,
a smeared crack damage model, with uniaxial tensile and compressive behaviour shown in Fig. 4, allows for the simulation of brittle
materials, like concrete and masonry.
According to this model, cracking is assumed to occur when the
stress reaches a critical failure surface, given by the relationship
between the equivalent pressure stress, p, and the Mises equivalent
deviatoric stress q, as it is illustrated in Fig. 5.

142

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

The basic unilateral contact law is described by the set of inequalities (1), (2) and by the complementarity relation (3):

h = u g 0 ==> h 0
n

Fig. 6. Stressdisplacement law of the smeared crack damage model representing


tension softening.

t 0

(3)

t n (u g ) = 0

(4)

Inequality (2) represents the non-penetration relation. Inequality (3) implements the requirement that only compressive stresses
(contact pressures) are allowed in each contact interface. Eq. (4)
is the complementarity relation which states that either separation with zero contact stress occurs or contact is realised with
possibly non-zero contact stress. For a discretised structure the
previous relations are written for every point of a unilateral
interface.
The behaviour in the tangential direction is dened by a static
version of the Coulomb friction law. In particular, two contacting
surfaces start sliding when the shear stress in the interface reaches
a critical value equal to:

t t = cr = |tn |

Fig. 7. An alternative continuum model: stressstrain behaviour of the masonry.

In the framework of this model, no individual macro cracks


are developed. On the contrary, the constitutive calculations are
performed independently at each integration point of the nite element model, thus the presence of cracks affects the stress and
material stiffness associated with each integration point. In addition, the compressive response of the material is modelled by an
elastic-plastic theory using a simple form of yield surface written
in terms of the equivalent pressure stress p, and the Mises equivalent deviatoric stress q (Fig. 5). The post-failure behaviour of the
damaged material is modelled with a tension stiffening law and
the stressdisplacement diagram shown in Fig. 6.
3.3.2. An alternative damage model for the arches
A slightly different continuum model is also used in the present
study. According to this model, the inelastic material behaviour of
masonry is also simulated by a cracking constitutive law for brittle materials. The damage is considered to be homogeneous within
an element even if the length of individual cracks is much smaller
than the element size. In addition, damage is described by microcracks oriented along mutually perpendicular planes. These cracks
are developed in the undamaged material when a maximum principal stress criterion is satised. Similar to the previously mentioned damage model, the presence of cracks affects the stress and
material stiffness associated with each integration point.
A tension softening model is considered for the stress and a
shear softening/retention law is used for the shear components of
stress. The crack model is described by the critical cracking stress
cr , the tension-softening modulus Es and the crushing strain
crush (Fig. 7). The shear retention factor is used to dene the residual shear stiffness for a cracked integration point in a cracking
analysis. This reduced shear modulus will have effect when the
normal stress across a crack becomes compressive.
3.3.3. Discrete model for the arches
For the verication of the results obtained by the usage of continuum models, a discrete model is also developed for a single arch
of the structure. The model consists of unilateral contact interfaces
standing for potential cracks, distributed in the body of the arch.
Along these interfaces, unilateral contact and frictional effects are
considered.

(2)

(5)

where tt and tn are the shear stress and the contact pressure
at a given point of the contacting surfaces respectively and is
the friction coecient. There are two possible directions of sliding
along an interface, so tt can be positive or negative depending on
that direction. Furthermore, there is no sliding if |tt | < |tn | (stick
conditions).
The Lagrange multiplier method is used to incorporate in the
equilibrium equations, the unilateral contact friction equations.
Finally, an alternative approach of a unilateral contact interface
with a non-zero tensile resistance is adopted in this study.
3.4. Two- and three-dimensional models
Several nite element models have been developed, for the simulation of the Cernadela Bridge. Four of them are used for the simulation of a single arch of the structure and a fth one for the investigation of the behaviour of the whole bridge. In particular, two
2d models are developed with the two mentioned damage laws,
one 3d model is also developed with the smeared crack damage
model in a single arch and one 2d discrete model is developed for
the same arch. Within the rst four models, a parametric investigation of the tensile strength of the masonry and of the width of the
arch has been considered. Finally, the proper material parameters
have been chosen for implementation on the whole structure.
In particular, the rst smeared crack concrete nite element
model is used for the simulation of the second arch (Arch 2) of the
Cernadela Bridge in two dimensions, Fig. 8. The main dimensions
of the bridge are given below [27]:
- Length of spans (right to left, downstream view): 3.58 m,
6.56 m, 10.01 m, 11.14 m, 10.30 m.
- Rise of arches (right to left, downstream view): 1.79 m, 3.77 m,
5.22 m, 5.80 m, 4.75 m.
The model consists of quadrilateral, four-node, plane stress elements with two translational degrees of freedom per node. A typical value for the length of each nite element is 0.03 m. A total
number of 4725 elements are used. In Fig. 9a the mesh of the arch
is shown.
For the implementation of the alternative damage model, the
mesh shown in Fig. 10 is used. The model consists of quadrilateral,
four-node, plane stress elements with two translational degrees of
freedom per node. A total number of 4351 elements are used. The
same mesh has been used in the framework of the model with
contact interfaces.

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

143

Fig. 8. Geometry of the Cernadela Bridge the simulated single arch.

Fig. 9. Mesh of the simulated single arch (a) two dimensional and (b) three dimensional model.

brick nite elements with three translational degrees of freedom


per node, are used for the mesh of the model. A number of 231,693
nite elements have been used. The inuence of loading in some
of the arches or the movement of abutments on the ultimate behaviour of the structure, are investigated. It is noted that the whole
structure was also simulated in [20], with a linear nite element
analysis model.

4. Inverse analysis and parameter identication

Fig. 10. Mesh of the simulated single arch two dimensional model of Marc.

A fourth, three dimensional nite element model has been developed, for the study of the same arch of the bridge, in three
dimensions. The width of the arch is considered equal to 0.5 m.
Three dimensional hexahedral nite elements with three translational degrees of freedom per node have been used. The total number of them is equal to 73,520, Fig. 9b. The smeared crack concrete
model is used for the investigation of the damage in this model. In
the described models, loading conditions include self-weight and a
concentrated load at the quarter span of the bridge.
In [27] the same arch of the structure was simulated with a
discontinuous nite element model, as well as with Ring 2.0 limit
analysis software [28]. In both models, a discrete modelling approach was considered, contrary to the present study where continuum damage models are mainly used. Consequently, comparison
between the results obtained from the continuum and discrete approach will be considered for this arch of the bridge. Thus, the ultimate (limit) load and the collapse mechanism received from both
approaches will be examined. This procedure is used for the validation of the parameters of the used damage models.
Finally, a three dimensional, continuum, nite element model
is developed, for the whole geometry of the bridge, Fig. 11. For the
investigation of the ultimate behaviour of the arch, the smeared
crack concrete model has been used. For the ll, a classical Mohr
Coulomb elastoplastic law has been chosen. Three dimensional,

Let us consider that a structure has suffered from a destructive


loading in the past. Usually small or larger deformations and damages or cracks remain after this experience. Using photogrammetry
we can accurately measure the current state of the structure. On
the other hand we can use the starting shape of the structure, if
available from drawings or estimates, and established mechanical
models in order to create loading and prediction scenarios on the
computer. Comparison of the predictions of these scenarios with
the measured state will allow us saying which one most probably occurred in the given structure. From the mathematical point
of view we formulate and solve a parameter estimation (identication) problem, or inverse problem, in order to adjust the parameters of the mechanical model so that existing damage or deformation patterns are reproduced with the highest accuracy [29]. We
can solve this problem by combining a parameterised mechanical
model with an optimisation algorithm, or use this formulation in
order to detect, at least, the existence of some defect within the
structure. The information of the surface measurements can be extended by either additional measurements or focused post processing of data in order to nd surface defects or cracks or by adding
information from the interior of the structure, by using for example geophysical prospection or other suitable techniques.
In this section, results related to the mechanical behaviour of
the Cernadela Bridge will be presented. For this reason, the mentioned models have been used in the framework of nite element
analysis. Within the rst smeared crack damage model, Youngs
modulus has been considered equal to 23 GPa, Poisson s ratio 0.2
and density 20 0 0 kg/m3 . The tensile strength of the structure is
considered equal to 0.5 MPa. Large displacement effects are neglected while the arch is considered to be xed to the ground.
NewtonRaphson incremental iterative procedure has been used
for the solution of the non-linear problem.

144

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

Fig. 11. Mesh of the whole Cernadela Bridge.

Fig. 12. Collapse mechanism obtained from the continuum nite element models (a) two dimensional and (b) three dimensional model.

Fig. 13. Collapse mechanism obtained from the discrete, two dimensional nite element model (a) separation stress = 0.25 MPa and (b) separation stress = 0.00 MPa.

By using the alternative nite element continuum model


the material parameters are considered as follows: Youngs
modulus, E = 23 GPa, Poissons ration 0.2, density 20 0 0 kg/m3 ,
critical cracking stress, cr = 0.25 MPa, tension-softening modulus,
Es = 2.5 GPa, crushing strain, crush = 0.003 and the shear retention factor equal to 1.0 (Fig. 7).
For the two dimensional discrete model, rst, no tension
strength is considered. Then, a low strength of tension equal to
0.25 MPa is used.
4.1. Parametric analysis of the single arch of the structure
Results obtained from the study of the second arch of the Cernadela Bridge will be presented in this section. For this reason,
continuum damage and discrete models have been developed in
two and three dimensions, respectively. The collapse mode and the
failure load received from these models, are compared with the
corresponding results obtained from a discrete modelling approach
presented in [27], in the same arch.

The failure mode which arises from the damage models is


the four hinges collapse mechanism, Fig. 12(a) and (b). The same
mechanism is received from the discontinuous nite element
model presented in [27]. A similar collapse mechanism is also obtained from the discrete model formulation which is used in this
article. Small differences about the location of the two hinges (at
the left side) which are presented are related with the value of
separation force, as it is shown in Fig. 13.
In addition, the inuence of the arch width is examined for
the two dimensional model considering values equal to 0.5 m
and 1.0 m. The corresponding results are found at the force
displacement diagrams shown in Figs. 14 and 15. In Fig. 16 are
summarised the forcedisplacement diagrams, obtained from the
damage models used in this article and the discrete models presented in [27]. According to these, the failure loads received from
the damage models are found between the corresponding values
of the discrete, three dimensional nite element model and the
limit analysis model with the Ring software, presented in [27]. As
the load is increased, the inclination of the diagrams change and

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

Fig. 14. Forcedisplacement diagrams for the second arch of the Cernadela Bridge
(width of the arch 0.50 m).

145

Fig. 15. Forcedisplacement diagrams for the second arch of the Cernadela Bridge
(width of the arch 1.00 m).

the diagrams tend to become horizontal, indicating that the structure is close to collapse. At the same time, the four hinges collapse
mechanism appears.

4.2. Study of the whole bridge with the damage model


In this section the mechanical behaviour of the whole Cernadela Bridge is under investigation. The same structure was considered as a unity in [1], thus no distinction between the arch and
the ll was taken into account. In the present study, a model in
which the archll interaction is taken into account is developed,
meaning that a different material law is applied to each of them.
A MohrCoulomb elastoplastic material law is adopted to simulate failure of the ll. In this framework, an angle of internal friction equal to 42 degrees and a cohesion equal to 0.5 MPa are chosen. The elasticity modulus of the ll is initially taken equal to
15 GPa. Then, a short parametric investigation of the inuence of
the material properties of the ll on the behaviour of the structure
is presented. The interaction between arch and ll is simulated by
a tie constraint, thus a condition which does not permit neither
sliding nor opening in the interface. At the end of this section, a
model with some pre-existing cracks, as they obtained from Photogrammetry (Fig. 1d), is presented.
The smeared crack damage model, which was previously presented, is used for the investigation of the failure of the arch. The
Youngs modulus has been considered equal to 23 GPa, Poissons
ratio 0.2 and density 20 0 0 kg/m3 . The same density and Poissons
ratio are taken also for the ll. The tensile strength of the arch
is considered equal to 0.5 MPa. Large displacement effects are neglected and the arch is considered to be xed to the ground. The
NewtonRaphson incremental iterative procedure has been used
for the solution of the non-linear problem.
The results which are presented in the following lines, demonstrate failure on the structure in case movement of abutments or
a static, trac load is applied to the bridge. In [20] a similar investigation was conducted by a linear nite element model, on the
same masonry arch. In [30] the inuence of the movement of abutments on the collapse mechanism of two dimensional stone arches
was investigated, by developing discrete nite element models.
Concerning the loading of the structure, three loading steps
have been developed. In the rst step the dead load of the bridge
is considered, while in the second step a uniformly distributed
load of 3 kN/m3 is applied to the structure. In the third step a

Fig. 16. Summary of the forcedisplacement diagrams for the second arch of the
Cernadela Bridge (width of the arch 0.5 m).

concentrate load or a movement of abutments is applied to the


bridge.
When a vertical displacement is applied to the fourth abutment of the structure, damage arises in the fourth and the fth
arch according to Fig. 17. Similarly, principle stresses of the linear
model presented in [20] become maximum in the same areas of
the fourth and fth arch. In addition, Fig. 18 shows the damage
of the fourth arch, in case a trac, static load is applied to it. A
close image is obtained by the linear model in [20], for the same
loading. In Fig. 19 the failure of the ll for the same case is given.
A comment related to the ultimate behaviour of the structure
can be made, in case the trac load of the fourth arch is accompanied with a transverse displacement of the fourth abutment (in a
direction vertical to the longitudinal axis of the bridge). According
to the upstream view shown in Fig. 20 the damage in this case is
expanded to the area of the spandrel walls of the fourth and fth
arch, contrary to Fig. 18 where damage arises almost exclusively in
the middle of the fourth arch. In addition, damage has been expanded to the fourth abutment, according to downstream view of
Fig. 21. This demonstrates that the behaviour of the structure is
signicantly inuenced in case a transverse loading is applied to
it, for instance after an earthquake.

146

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

Fig. 17. Damage on the fourth and fth arch for a vertical displacement of the fourth abutment.

Fig. 18. Damage of the arch for a trac load in the fourth arch.

Fig. 19. Damage of the ll for a trac load in the fourth arch.

Fig. 20. Damage for a trac load in the fourth arch and a transverse movement in the fourth abutment upstream view.

Fig. 21. Damage for a trac load in the fourth arch and a transverse movement in the fourth abutment downstream view.

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

147

Fig. 22. Crack opening on the fth arch for a vertical displacement of the fourth abutment Fill Youngs modulus = 23 GPa.

Fig. 23. Crack opening on the fth arch for a vertical displacement of the fourth abutment Fill Youngs modulus = 8 GPa.

The mentioned model is nally enhanced by taking into account pre-existing cracks, as they were obtained from the Photogrammetry (Fig. 1d). By taking into account the current state of
damage and deformation, the estimation of the real structural behaviour and strength of the arch can be more realistic and accurate.
To simulate the cracks, a unilateral contact law has been applied to the cracked surfaces. Thus, zero tensile resistance is given
in each interface. In addition, three parametric simulations of the
ll have been taken into account. First, the initial Youngs modulus of the ll (15 GPa) and a vertical displacement of the fourth
abutment are considered. Then, the Youngs modulus is considered
equal to 8 GPa and 23 GPa, respectively. The goal of this investigation is to examine the mechanical behaviour of the structure
with the cracks as well as to understand the inuence of the lls
Youngs modulus on the behaviour of the structure.
According to the output, the bigger elasticity modulus of the
ll results in bigger crack openings (Fig. 22), in comparison with
the case with smaller elasticity modulus (Fig. 23), for the same
value of movement of the abutment. This shows that the properties of the ll may inuence the ultimate behaviour of the cracked
structure.

Fig. 24. Crack opening abutments vertical displacement diagram for a variation
of the elasticity modulus of the ll.

Additionally, the diagrams shown in Fig. 24 indicate non-linear


response as well as bigger crack openings for bigger Youngs modulus of the ll.

148

M.E. Stavroulaki et al. / Advances in Engineering Software 101 (2016) 136148

5. Conclusions
In the present study terrestrial photogrammetry is used for geometry reconstruction of a real masonry arch bridge located in
Spain, the metric model is used for subsequent implementation
in structural assessment tasks. Terrestrial photogrammetry significantly contributes to the accurate geometric representation of historical structures with milimetric precision. It is important to note
some limitations on the use of the photogrammetric method. Although the high level of detail recorded, this may be not enough
for detection of movement or subtle displacements of masonry
blocks. The main advantage of such surveying method is based
on the accuracy reached for the positioning of real masonry elements that will subsequently support the accurate geometric
characterisation during the nite elements based model of the
whole structure. The exact geometry obtained from this method
is then used for the investigation of the ultimate behaviour of the
structure.
In particular, several two and three dimensional, non-linear
models have been developed. To depict the limit load and the collapse mechanism of the structure, continuum damage models have
been used and compared with discrete approaches conducted in
the present as well as in older studies.
According to the results, the classical four hinges mechanism
arises from the structural analysis of the bridge. Moreover, the inuence of parameters such us the width of the structure and the
tensile strength of the material in the forcedisplacement diagrams
is shown.
The simulation of the whole structure demonstrates that a possible out of plane movement of abutments will cause signicant
damage to the bridge. Finally, useful results are obtained when
the existing cracks of the structure are incorporated in the simulation. Values of crack opening and the pattern of hinges which are
activated are received after simulation. Additionally, the variation
of the elasticity modulus of the ll may signicantly inuence the
values of crack opening in the structure.
Acknowledgements
The work of Dr. Georgios Drosopoulos is being supported
by a research project implemented within the framework of
the Action Supporting Postdoctoral Researchers of the Operational Program "Education and Lifelong Learning" (Actions Beneciary: General Secretariat for Research and Technology), and conanced by the European Social Fund (ESF) and the Greek State.

References
[1] Stavroulaki ME, Riveiro B, Drosopoulos GA, Arias P, Stavroulakis GE. Integrated
modeling and evaluation of Masonry Bridges using terrestrial photogrammetry. In: Topping BHV, Ivnyi P, editors. Proceedings of the twelfth international conference on computational structures technology, Stirlingshire, UK.
Civil-Comp Press; 2014 Paper 242 doi:10.4203/ccp.106.242.
[2] Rekik A, Lebon F. Homogenization methods for interface modeling in damaged
masonry. Adv Eng Softw 2012;46:3542.

[3] Sacco E. A nonlinear homogenization procedure for periodic masonry. Eur J


Mech A/Solids 2009;28(2):20922.
[4] Massart TJ, Peerlings RHJ, Geers MGD. Structural damage analysis of masonry
walls using computational homogenization. Int J Damage Mech 2007;16:199
226.
[5] Massart TJ, Peerlings RHJ, Geers MGD. An enhanced multiscale approach for
masonry wall computations with localization of damage. Int J Numer Methods
Eng 2007;69:102259.
[6] Formisano A, Florio G, Landolfo R, Mazzolani FM. Numerical calibration of
an easy method for seismic behaviour assessment on large scale of masonry
building aggregates. Adv Eng Softw 2015;80:11638.
[7] Sarhosis V, Oliveira DV, Lemos JV, Lourenco PB. The effect of skew angle on the
mechanical behaviour of masonry arches. Mech Res. Commun 2014;61:539.
[8] de Felice G. Assessment of the load-carrying capacity of multi-span masonry
arch bridges using bre beam elements. Eng Struct 2009;31:163447.
[9] Ramosa LF, De Roeck G, Loureno PB, Campos-Costa A. Damage identication
on arched masonry structures using ambient and random impact vibrations.
Eng Struct 2010;32:14662.
[10] Betti M, Vignoli A. Numerical assessment of the static and seismic behaviour of the basilica of Santa Maria allImpruneta (Italy). Constr Build Mater
2011;25:430824.
[11] Ferris MC, Tin-Loi F. Limit analysis of frictional block assemblies as a
mathematical program with complementarity constraints. Int J Mech Sci
2001;43:20924.
[12] Fishwick RJ, Liu XL, Begg DW. Adaptive search in discrete limit analysis problems. Comp Methods Appl Mech Eng 20 0 0;189:93142.
[13] Milani G. Upper bound sequential linear programming mesh adaptation
scheme for collapse analysis of masonry vaults. Adv Eng Softw 2015;79:91
110.
[14] Remondino F, Girardi S, Rizzi A, Gonzo L. 3D modeling of complex and detailed
cultural heritage using multi-resolution data. J Comput Cult Herit 2009;2(1) 2.
[15] Jiang R, Jauregui DV. Development of a digital close-range photogrammetric
bridge deection measurement system. Measurement 2010;43:14318.
[16] Solla M, Lorenzo H, Riveiro B, Rial FI. Non-destructive methodologies in
the assessment of the masonry arch bridge of Traba, Spain. Eng Fail Anal
2011;18:82835.
[17] Riveiro B, Solla M, de Arteaga I, Arias P, Morer P. A novel approach to evaluate masonry arch stability on the basis of limit analysis theory and nondestructive geometric characterization. Autom Constr 2013;3:1408.
[18] Arias P, Ordonez C, Lorenzo H, Herraez J, Armesto J. Low-cost documentation
of traditional agro-industrial buildings by close-range photogrammetry. Build
Environ 2007;42:181727.
[19] K.J. Sandmeier, ReexW manual v.4.5. (from http://www.sandmeier-geo.de).
Sandmeier Scientic Software, 2007.
[20] Lubowiecka Arias P, Riveiro B, Solla M. Multidisciplinary approach to the assessment of historic structures based on the case of a masonry bridge in Galicia (Spain). Comput Struct 2011;89:161527.
[21] Drosopoulos GA, Stavroulakis GE, Massalas CV. Limit analysis of a single
span masonry bridge with unilateral frictional contact interfaces. Eng Struct
2006;28:186473.
[22] Leftheris BP, Stavroulaki ME, Sapounaki AK, Stavroulakis GE. Computational
mechanics for heritage structures. Southampton (UK): WIT Press; 2006.
[23] Stavroulaki ME, Stavroulakis GE. Unilateral contact applications using FEM
software. In: Sofonea M, Viano JM, editors. Special Issue on Mathematical
Modeling and Numerical Analysis in Solid Mechanics, Guest Editors. Int J Appl
Math Comput Sci 2012;12(1):10111.
[24] Cervera M, Pela L, Clemente R, Roca P. A crack-tracking technique for localized
damage in quasi-brittle materials. Eng Fract Mech 2010;77:243150.
[25] Anthoine A. Derivation of the in-plane elastic characteristics of masonry
through homogenization theory. Int J Sol Struct 1995;32:13763.
[26] Campo M, Drosopoulos GA, Stavroulakis GE. Unilateral contact and damage
analysis in masonry arches. Vietnam J Mech 2009;31:18590.
[27] Riveiro B, Caamao JC, Arias P, Sanz E. Photogrammetric 3D modelling and
mechanical analysis of masonry arches: an approach based on a discontinuous
model of voussoirs. Autom Constr 2011;20:3808.
[28] Ring 2.0 Theory & Modelling Guide, LimitState. Sheeld, UK. Available at: http:
//www.limitstate.com/ring.
[29] Stavroulaki ME, Drosopoulos GA, Stavroulakis GE. Unilateral analysis and damage identication in masonry structures. In: Topping BHV, Tsompanakis Y, editors. Proceedings of the thirteenth international conference on civil, structural
and environmental engineering computing. Stirlingshire, UK: Civil-Comp Press;
2011 Paper 61.
[30] Drosopoulos GA, Stavroulakis GE, Massalas CV. Inuence of the geometry and
the abutments movement on the collapse of stone arch bridges. Constr Build
Mater 20 08;22:20 010.