You are on page 1of 8

23rd International Congress on Sound & Vibration

Athens,

Greece

10-14 July 2016

ICSV23

DEVELOPMENT OF STRUCTURAL RAILWAY MONITORING


SOLUTIONS USING FBG SENSORS
Georges Kouroussis, Damien Kinet, Vronique Moeyaert, Christophe Caucheteur
University of Mons UMONS, Faculty of Engineering, 7000 Mons, Belgium
email: georges.kouroussis@umons.ac.be

Julien Dupuy
Multitel, 7000 Mons, Belgium

Structural health and operation monitoring are of growing concerns in the development of railway
networks. Conventional systems of infrastructure monitoring already exist (e.g. axle counters,
track circuits) but present some drawbacks. However, the use of optical fibre sensors, and more
particularly fibre Bragg grating (FBG) sensors, becomes a realistic alternative due to their immunity to electromagnetic fields and simple multiplexing. The paper analyses the implementation of
FBG sensors as vibration sensors along a rail in order to obtain usable signals of sufficient interest
for weigh-in-motion and axle counting. This study is completed by a numerical analysis in order
to better discern and understand the location and the orientation of these sensors from the stress
transfer caused by the train passage to the track. Different data processing methods were tested
to estimate the train speed and weight. The results obtained are compared to strain gauges data
as well as train manufacturer data (for train axle loads) and video recordings (for train speed).
A good agreement is observed between these different results, demonstrating the applicability of
FBG sensors to monitoring applications. A particular attention is also paid to the advantages of
these new sensors.

1.

Introduction

Recently, a large amount of research in the structural monitoring domain has been undertaken,
particularly since the widespread development of high-speed rail lines. Furthermore, the interest of
scientific and technical communities continues to grow, with the development of new techniques,
either intrusive or non-intrusive, that use several sensor technologies. Compared to conventional
techniques such as axle counter sensors and the track circuits, vibration-based monitoring systems
emerged the last years, including fibre technology. They proved their efficiency in various civil engineering structures and structural monitoring [1]. However, there is a lack of information about the
sensor positioning since mechanisms that contribute to the track dynamics are complex [2].
The present paper focuses on the use of fibre Bragg gratings (FBG) as reliable and robust sensors
for railway monitoring. After a brief description of fundamentals of FBG sensors, a study of stress
transfer from the train passage to the track using predictive numerical models is presented. These
models are used to validate experimental devices and to determine the best locations of FBG sensors.
A test zone is then presented where various FBG sensors have been placed in order to have information
about the train speed and the weigh-in-motion. Various signal processing techniques are employed to
determine these data.
1

The 23rd International Congress of Sound and Vibration

2.

Principle of FBG sensors

Sensors based on wavelength modulation mainly rely on FBGs. A Bragg grating consists of a
periodic modification of the core refractive index. It is realized by subjecting the fibre core to an UV
interference pattern. The effective refractive index of the core before the grating is made is called
neff . The main idea is to deform the grating in order to let the pitch length (Figure 1) and the index
neff vary, leading to a variation of the Bragg wavelength Bragg . The Bragg wavelength Bragg can be
expressed by
Bragg = 2neff .

(1)

The wavelength Bragg shift changes linearly with both strain and temperature. When the grating
part is subjected to external disturbance, the period of the grating will be changed and the Bragg
wavelength will vary accordingly.
Generally, optical fibres are cylindrical silica waveguides made of two concentric layers, the core
and the cladding, which guide light thanks to a slight refractive index difference between both materials. A fibre-based sensor system consists of a light transmitter, a receiver, an optical fibre, a modulator
element, and a signal processing unit. In the case of FBG sensors, the useful information is contained
in the Bragg wavelength shift. The latter can be measured by different demodulation techniques.
One of the most straightforward and cost-effective consists in turning the wavelength shift into an
amplitude change, by using an optical fibre. This is the so-called edge filter technique, controlled the
interrogation unit (often called interrogator). The output signal is therefore a voltage proportional to
the deformation along the fibre at the FBG position (strain sensor). The sensitivity of FBG sensor
can therefore be expressed in [pm/] and it depends on many factors (core geometry and material,
Bragg value, . . . ).
Ii

It

Incident spectrum
Broadband
light source

Transmitted spectrum
Detector

Ir

Cladding

Bragg

Fibre Bragg grating (FBG)

Core

Reflected spectrum

Figure 1: Measurement principle of FBG sensor [3].


FBG sensors are very promising vibration sensors in the railway area as they have multiple advantages compared with the conventional methods that already have been used [4]. The most important
benefit is that they are electromagnetic immune, can be sorted in an extended array of sensors (quasidistributed sensors) that will cover a very long distance of the rail monitoring with one single fibre
and one interrogator. This task is however complicated and a deep analysis should be made before we
would be able to conclude to a system that will operate the railway monitoring in real time and will
provide robust and reliable information.
2

ICSV23, Athens (Greece), 10-14 July 2016

The 23rd International Congress of Sound and Vibration

3.

Finite element model for calculating the stress into the rail during the
passage of trains

In order to find the optimal and reliable position for strain measurement, various numerical simulations have been performed. This virtual prototyping step aims to primary provide information
concerning the axle counting, the train speed and the calculation of the axle loads. The proposed
simulations and the corresponding results are based on Abaqus software and more particularly on
the Abaqus/Standard module. The following definitions and assumptions have been chosen for the
simulation:
A three-dimensional model of the rail is defined, according to its geometry (e.g. 50E2 rail in
the proposed case see Fig. 2(a)).
The mechanical behaviour of the rail is assumed to be linear. The properties of the steel
(Youngs modulus 210 GPa; density 7800 kg/m3 ; Poissons ratio 0.3) are used in all cases.
The boundary conditions are chosen according to two studied cases:
for simulations related to three point flexural tests (laboratory tests), the rail length is fixed
(1 m) with revolute joints defining the two supports;
for simulations related to railway conditions, a finite length of the rail is chosen (10 m,
sufficiently large to represent accurately the local deformation between two sleepers).
In the case of simulation related to railway conditions, as a matter of fact, the length of the
flexible part of the track must be sufficient to embrace the rail deflection. This corresponds in
most practical cases to a minimum of ten sleepers, beyond the region of interest. Considering a
complete vehicle, its length can be superior to the length so that a rigid track is added on both
sides of the finite element model of the track. Indeed, in some situations, the vehicle length can
be important and a bigger track length imposes an excessive and useless number of degrees of
freedom for the track simulation. The use of a smaller length of the track allows a focus on
the region of interest of railway-induced deformation. Each support is defined by an equivalent
vertical stiffness (like a Winkler foundation) representing the overall stiffness of the railpad,
of the ballast and of the foundation (Fig. 2(b)). The model has been compared to numerical
solutions obtained by a compound multibody/finite element approach [5].
The loading is defined vertically according to the expected degree of accuracy. Firstly, a simple
load is defined on a node of the rail; secondly, a set of loadings define the entire train; lastly, a
set of loaded rigid wheels reproduce the effect of moving train with specific definition for the
contact (tie contact).
The simulation step is based on static and dynamic calculations. As expected, the nominal train
speeds are sufficiently low compared to the wave propagating into the rail so that the quasistatic effect dominates the rail response. Therefore, a dynamic analysis is not necessary in all
cases. In order to include the successive position of the wheels, the moving force option is
used: a path is defined along which move the wheels and a number of frames is defined by the
user in order to ensure the right position of each contact.
For the rail, a 20-node quadratic brick (C3D20R) is used for the meshing. It is a general purpose
quadratic brick element, with reduced integration (222 integration points). The shape functions and the integration scheme are well adapted for flexural motion. The element behaves very
well and is an excellent general purpose element. It also performs well for isotropic material
behaviour and in bending and rarely exhibits hourglassing despite the reduced integration.
Figure 3 compares the numerical results and their experimental counterpart for the case related to
three point flexural tests (laboratory tests). FBG sensors have been placed on the rail foot along
a direction parallel to the rail. This static test allows the confirmation of the sensor sensitivity
(1.21 pm/). A very good agreement is observed, validating the numerical model for advanced
simulation (railway configuration). Although not presented in this paper, these expected results also
demonstrated that the rail monitoring using strain measurement is very accurate for the calculation
ICSV23, Athens (Greece), 10-14 July 2016

The 23rd International Congress of Sound and Vibration


73.7
(Er ,Ir ,r ,Ar )

x
151

15

140
(a) Cross-section of a
50E2 rail

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

(b) Modelling of the track

Figure 2: Position of the problem and finite element simulation configuration.


of the train speed and the wheel positions especially when the FBG sensors are located on the rail
feet. Furthermore, the results of the measurement on the neutral axis can provide sufficient accuracy
compared with the first one but they are able to calculate the contact position. This gives an advantage
because this cannot only provide information about the static load of the trains but it can also identify
possible local defect on the train wheels (presence of wheel flats).
P

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxx
xxxxx
xxxxx
xxxxx
xxxxx
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx

quasi-distributed FBG sensors

Strain [m/m]

0.4
0.2
0
0.2

Finite element model


Measurement on laboratory

0.4
0

0.2

0.4

0.6

0.8

Location along the rail [m]

Figure 3: Strain along the rail specimen in the laboratory test in comparison with the finite element
model results.

4.

Field experimental setup

4.1

Selected site

A site was chosen in Belgium for the measurement campaign. Several sensors were placed on a
track close to a railway station (Fig. 4):
Various FBG sensors have been placed on the rail web and foot. Four of them have been glued
on the web, on the neutral axis, and oriented at 45o (sensors a b c d, see Fig. 5). Two other
FBG sensors have been glued on the rail foot, along the rail length (sensors e f ) and spaced
with the distance L, as illustrated in Fig. 5.
4

ICSV23, Athens (Greece), 10-14 July 2016

The 23rd International Congress of Sound and Vibration

Four strain gauges have been fixed on the web on the same rail but at the opposite side of the
FBG sensors a b c d, using a similar configuration. This set of sensors offers the way to
compare both FBG and gauge sensors.
Several classical accelerometers have been used for additional measurement on the track and
on the trackbed. For the comparison of FBG sensors, two vertical accelerometers fixed on two
adjacent sleepers and one vertical accelerometer on the rail foot have been used.

strain gauges

accelerometer on rail

FBG sensors

accelerometers on sleepers

Figure 4: Selected site and field instrumentation.

P (x, t)
c
d

a
b

is
ax
al
e
utr
e
n

Figure 5: Wheel/rail vertical force and track deflection estimation using FBG sensors.

4.2

Signal processing

The positioning of sensors on the neutral axis is the best compromise between sensitivity, crosstalk
and influence length to have an accurate description of the shear strain a b (or c d ). One full
bridge set-up made with the four sensors easily leads to the desired output between brackets in Eq. (2).
The sensors configuration is shown in Figure 5 where four gauges a b c d are placed on the neutral
axis of the track web to determine the vertical loading P . If i is the strain measured by a sensor and
if there is no support (sleeper) under the rail within this gaged region, the vertical wheel/rail force is
calculated using
P (x, t) =

EIt
(a b + c d )
(1 + )H

(2)

where E is the Youngs modulus, the Poissons ratio, t the web thickness, I the geometrical moment
of inertia and H first moment of area.
ICSV23, Athens (Greece), 10-14 July 2016

The 23rd International Congress of Sound and Vibration

The placement of sensors on the rail foot offers a way to measure the track deflection. From the
beam theory, it is relatively simple to demonstrate that maximum bending strain p is proportional
to the rail deflection w and, using the Euler formulation to represent the behaviour of the track, the
theoretical rail deflection can be obtained from the theory of beam on elastic foundation [6]
w(x, t) = w(x vt) =
q

P
e|xvt| [cos(|x vt|) + sin(|x vt|)] ,
8EI 3

(3)

where = 4 4EIf is introduced, representing a ratio of flexibility between the foundation and the
rail. The track (two parallel rails with periodically fastened sleepers) is therefore treated as an Euler
Bernoulli elastic beam lying on a Winkler foundation, defined by its stiffness Kf per unit of length,
including railpad, ballast and foundation contributions. From the accelerometer placed on the rail, this
deflection can be obtained through 2 successive integrations combined to specific signal processing
techniques. Rail deflections, often called quasi-static defections, take into account the number nc of
carriages for each studied vehicle. Periodicities of wheelsets, bogies and carriages, defined as [7]
v
La
v
fb =
Lb
v
fc =
,
Lc

fa =

(4)
(5)
(6)

depend on the wheelset spacing La , bogie spacing Lb and carriage length Lc and where v is the vehicle
speed (assumed to be constant in the frequency analysis).
Regarding the use of FBG sensors e and f , an interesting and robust method [8, 9], called train
speed calculation using dominant frequency, is based on the isolation of frequencies fc,n in the track
vibration spectrum. Indeed, the passing of axle load with periodicity generates significant peaks in the
measured spectrum: at a specific location in the track, a train consists of a number of similar events
with certain delay times. Each peak is regularly spaced by
fc =

v
.
Lc

(7)

The calculation technique is defined in three steps where an initial speed is required, and is summarized as follow.
In order to better estimate the fundamental dominant excitation frequency, a cepstral analysis is
applied to the track vibration signal spectrum Wi (f ):
Cw ( ) = iDFT (log |Wi (f )|) .

(8)

This operation is able to reveal and quantify fc (the carriage excitation frequency) but also fa
(the wheelset excitation frequency) and fb (the bogie excitation frequency). All these values are
dependant of the vehicle speed and the main vehicle geometry.
The use of the dominant frequency method is based on the running rms instead of the original
signal
s Z
tt0
1 t0 2
wrms, (t0 ) =
(9)
wi (t)e dt .
t0
This allows a clearer visualization of passing of each wheelset.
To avoid limitations due to the signal frequency resolution and the other sources of excitation, a
regression analysis is performed which combines the ground vibration measurements with the
analytical solution based on Eq. (3), taking into account all the axle loads.
6

ICSV23, Athens (Greece), 10-14 July 2016

The 23rd International Congress of Sound and Vibration

Correlation functions can also be used for FBG sensors e and f to describe the average relation
between the two variables. The cross-correlation is a function only of the delay and the maximum
value does not necessarily occur at t = 0. The cross-correlation function can be approximated by the
time average:
Z
1 T
Ref ( ) = lim
e (t)f (t + )dt .
(10)
T T 0

Knowing the distance L between the two sensors e and f , the train speed can be calculated as v = L .
4.3

Results

Acceleration [m/s2 ]

60
Force [kN]

10

using strain gauges


using FBG sensors

80

40
20

on rail foot
on sleeper (left)
on sleeper (rail)

0
0

4
Time [s]

10 0

(a) Measuring weight

4
Time [s]

(b) Measuring acceleration

Figure 6: Example of data recorded with strain gauges and FBG sensors (a), and accelerometers (b).
Figure 6 shows typical results obtained from the various devices installed on the studied track.
A comparison between strain gauges and FBG sensors signals is presented in Figure 6(a), after unit
conversion using Eq. (2). A very good agreement is observed between the two curves, especially
when a non-zero strain is recorded, validating the measurement using FBG sensors. Note that the
results are related to one rail (and the correct axle load on the entire vehicle must be multiplied by 2).
The results confirm the usability of FBG sensors for calculating the axle loads passing on a small area
of a track.
Table 1: Train speed calculation Comparison between vibration measurement methods and video
frame counter.
Train
#1
#2
#3
#4

estimated speed
estimated speed
estimated speed
using dominant frequency using correlation using video recording
55.3 km/h
76.1 km/h
29.6 km/h
59.4 km/h

54.5 km/h
72.8 km/h
30.3 km/h
58.4 km/h

55.4 1.8 km/h


72.8 2.4 km/h
29.3 0.9 km/h
58.6 1.9 km/h

Figure 6(b) shows the results of acceleration signals obtained using classical accelerometers. Obtained levels on rail and on sleepers are of the same order of magnitude and the passing of each
wheelset is clearly visible on the curve, so as validating both sensor locations. Table 1 provides the
ICSV23, Athens (Greece), 10-14 July 2016

The 23rd International Congress of Sound and Vibration

information related to the speed obtained by the train speed calculation using dominant frequency
and using the correlation functions. Again it is noticeable that there is a good agreement between the
various methods for estimating the vehicle speed. The advantage of the calculation using dominant
frequency is its robustness against external perturbation like noise or the location of the sensors [9].

5.

Conclusion

Some important requirements for classical sensors (strain gauges) have been used as a guidance for
the positioning of fibre-based sensors and the analysis of resulting signals. Position and orientation
can be borrowed from the experience of strain gauges and applied to fibre sensor measuring the
local deformation. The knowledge of track dynamics allowed the understanding of some specific
phenomena and various simulations using finite element model validated the positioning of FBG
sensors. Appropriate signal processing methods are then presented and compared between them. It
turns out that FBG sensors are a promising way for railway monitoring.

Acknowledgements
This work is financially supported by the INOGRAMS (Innovations for a Global Rail Management System) project (convention 7171) of The Wallonia (Belgium) in the frame of the LOGISTICS
IN WALLONIA competitiveness cluster. Christophe Caucheteur is an F.R.S.-F.N.R.S. Research Associate and recipient of an ERC Starting Independent Researcher Grant (grant agreement No 280161
called PROSPER). The authors thank the partners (Alstom and Infrabel) for the fruitful discussions.

REFERENCES
1. Kinet, D., Mgret, P., Goossen, K. W., Qiu, L., Heider, D. and Caucheteur, C. Fiber bragg grating sensors
toward structural health monitoring in composite materials: Challenges and solutions, Sensors, 14 (4), 7394
7419, (2014).
2. Connolly, D. P., Kouroussis, G., Laghrouche, O., Ho, C. and Forde, M. C. Benchmarking railway vibrations
track, vehicle, ground and building effects, Construction and Building Materials, 92, 6481, (2015).
3. Ye, X. W., Su, Y. H. and Han, J. P. Structural health monitoring of civil infrastructure using optical fiber
sensing technology: A comprehensive review, The Scientific World Journal, Article ID 652329, 11 pages,
(2014).
4. Kouroussis, G., Caucheteur, C., Kinet, D., Alexandrou, G., Verlinden, O. and Moeyaert, V. Review of
trackside monitoring solutions: From strain gages to optical fibre sensors, Sensors, 15 (8), 2011520139,
(2015).
5. Kouroussis, G. and Verlinden, O. Prediction of railway induced ground vibration through multibody and
finite element modelling, Mechanical Sciences, 4 (1), 167183, (2013).
6. Timoshenko, S., Strength of materials, Van Nostrand, New York, USA (1942).
7. Kouroussis, G., Connolly, D. P. and Verlinden, O. Railway induced ground vibrations a review of vehicle
effects, International Journal of Rail Transportation, 2 (2), 69110, (2014).
8. Kouroussis, G., Connolly, D. P., Forde, M. C. and Verlinden, O. Train speed calculation using ground
vibrations, Journal of Rail and Rapid Transit, 229 (5), 466483, (2015).
9. Kouroussis, G., Connolly, D. P., Laghrouche, O., Forde, M. C., Wookward, P. and Verlinden, O. Robustness
of railway rolling stock speed calculation using ground vibration measurements, MATEC Web of Conferences, 20, 07002, (2015).

ICSV23, Athens (Greece), 10-14 July 2016

You might also like