GPS MONITORING OF THE FATIH SULTAN MEHMET SUSPENSION BRIDGE BY
USING ASSESSMENT METHODS OF NEURAL NETWORKS
O. Akyılmaz“*, R. N. Celik*, N. Apaydın ^ T. Ayan*
* ITU Civil Engineering Faculty, 34469 Maslak Istanbul, Turkey — (akyilma2, celikn, ayan)@itu.edu.tr
^ 17. Regional Headquarters of Highways, Zincirlikuyu Istanbul, Turkey — apaydin@boun.edu.tr
KEY WORDS: Geodesy, Monitoring, Modelling, Artificial Neural Networks, Prediction
ABSTRACT:
The second suspension bridge connecting the continents Asia and Europe, namely, Fatih Sultan Mehmet Bridge, has been monitored
by using GPS technique. For this end permanent GPS observations with 0.1 seconds epoch interval were recorded for the same days
of consecutive weeks. In addition to GPS observations, some other data belong to influencing factors such as traffic volume and
weather conditions for the corresponding observation time were collected. At first step the time series of the respective point
component displacements (deformations) were composed and linked to the data such as time, traffic volume and weather conditions.
Then a detailed comparison of the individual observation days was investigated. Further on, an artificial neural network, from the
family of soft computing methods is adapted in order to describe the deformation processes with respect to influencing factors. Such
studies have been of special interest after the 17 August Earthquake in North Anatolian Fault Zone (NAFZ) since new earthquakes
are expected. Therefore, monitoring of big engineering structures like bridges will bring important information for disaster
management and risk analysis. The results present that artificial neural networks are efficient tools for modelling complex
behaviours of deforming objects regarding the causing factors especially in case of continuous monitoring systems.
1. INTRODUCTION
Monitoring of engineering structures has become of importance
particularly after the possibility of destructive natural
catastrophes has been assumed to be increased. For this end, big
engineering structures like suspension bridges, viaducts, tunnels
and high buildings etc. have been subjected to continuously
monitoring surveys. The technological developments in high
precision point positioning systems together with no-human
data transmission techniques without any atmospheric
obligation have led to easily adapting such monitoring systems
for the objects in question.
Fatih Sultan Mehmet Bridge is the second suspension bridge
connecting the Asia and Europe. The construction has been
completed in 1987 and since July 1988, it served as the second
connection between Anatolian and European side for the
Istanbul dwellers. Daily, an average of sixty thousand vehicles
including automobiles, motorbikes, long vehicles, buses,
minibuses and trucks pass over the bridge. This number shows
how frequent the bridge is used. Therefore, any disaster which
may ruin the bridge will not only bring structural loss but also
many people will be damaged or even died.
It has long been a problem to geodesists to find efficient
solutions to approximate functions that define geodetic
deformations, especially when dealing with continuously
monitored processes. A deforming object can be considered as a
dynamic system (Pfeufer 1994, Welsch 1996, Heunecke and
Pelzer 1998, Miima and Niemeier 2004) whereby, forces acting
on the object (both internal and external loads) are regarded as
input signals that lead to geometrical changes e.g.
displacements and distortions as output signals. In most cases,
mathematically description of a dynamic deformation process is
very complex and using deterministic functions is not adequate
to depict the behaviour of the deforming object. Up to now,
many different methods were developed, it is however
generally agreed upon that, there exist no single method that
can satisfactorily describe the structural deformation as its
underlying processes are normally so complex to be expressed
by one simple expression.
The present study motivates the use of artificial neural networks
for modelling the behaviours of deforming objects regarding the
causing effects such as atmospheric conditions, traffic volume.
Artificial neural networks are inspired from biological systems
in which large numbers of neurons, which individually perform
rather slowly and, imperfectly, collectively perform
extraordinarily complex computations that even the fastest
computers may not match. This new field of computing method
is recently widely used by different disciplines such as
prediction and control engineering, image processing and
identification, pattern recognition, robotic systems etc. It is very
efficient tool for complex system identification in general.
2. STRUCTURAL DEFORMATION AS A DYNAMIC
SYSTEM
A dynamic system, in general, is characterized by input signals,
including all possible influences acting on the object leading to
the output signals. In case of structural deformation, acting
forces are regarded as input signals whereas the resulting
changes in the coordinate components are output signals (Fig.
1).
Input Output
signals signals
Pr essure X
Temperature .
Humidity — —», Bridge [Dunes Y
Wind Speed
Traffic Volume h
702
Inter
Fig
Heur
syste
or wl
If th
be t]
the o
outpu
equat
mode
priori
are tl
black
Artifi
mode
doma
input
make
transí
descr
mode
Artifi
regarc
neuro
have
produ
systen
compl
quanti
systen
a mai
Indeec
artific
devek
of or
compi
an iter
perfor
2 depi
netwoi
a real
comm
gros:
Inpu
x -
X3 —
The fui
