See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/252389017

Track Circuit automatic diagnosis based on a local electrical modelling

Article · January 2006

CITATIONS READS
5 388

4 authors, including:

Latifa Oukhellou Patrice Aknin

Institut Français des Sciences et Technologies des Transports, de l’Aménagement et … IRT System X
115 PUBLICATIONS   1,211 CITATIONS    125 PUBLICATIONS   824 CITATIONS   

SEE PROFILE SEE PROFILE

Thierry Denœux
Université de Technologie de Compiègne
281 PUBLICATIONS   6,367 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

VirMaLab View project

Asset management for Railways View project

All content following this page was uploaded by Thierry Denœux on 02 October 2014.

The user has requested enhancement of the downloaded file.

 Track Circuit automatic diagnosis based on a local electrical modelling
1,2,4 1,3 1 4
A. Debiolles , L. Oukhellou , P. Aknin , T. Denoeux

1
INRETS, Arcueil, France; 2SNCF, Paris, France; 3CERTES - Université Paris 12, Créteil, France;
4
UTC Heudiasyc UMR CNRS 6599, Compiègne, France

Abstract
The ground-to-train transmission system (TVM) is a vital factor in railway safety on high speed lines. It is
used to continuously transmit signalling information to the driver in the cab. The ground-based part of the
system consists of a compensated UM71-type track circuit including a transmitter, a receiver and
trimming capacitors, whose characteristics may change because of ageing or track maintenance
operations. This can lead to serious defects and even interrupt the TVM. It is therefore important to detect
the defects as early as possible to maintain the system safe and available. In this paper, we propose a
diagnosis method based on a local electrical modelling of the track circuit. The TVM carrier signal is
simulated, and the parameters of the local model are optimized so that the simulated signal is as close as
possible to the real measured one, recorded by an inspection vehicle. It makes it possible both to detect
the track circuit defects, and to estimate their graveness. We detail more precisely the diagnosis of
trimming capacitors.

1. Introduction

The ground-to-train transmission system (TVM) is a very important factor in railway safety on high speed
lines. It is used to continuously transmit signalling information to the driver in the cab. The groung-based
part of the system consists of a compensated UM71-type track circuit including a transmitter, a receiver
and trimming capacitors, whose characteristics may change, because of ageing or track maintenance
operations. This can lead to serious defects and even interrupt the TVM, leading to the train’s stop. It is
therefore important to detect the defects of this device as early as possible to maintain the system’s
safety and availability. To assess the availability of the track circuits, inspection cars regularly record the
TVM carrier signal (called I cc). Today, those signals are analysed mainly manually, only in a corrective
maintenance context. The aim of our work is to automate this analysis, both to accelerate it and to detect
growing defects, in a context of predictive maintenance. In this article, we present a diagnosis method
based on a local electrical modelling of the track circuit, dedicated to detect the trimming capacitors
defects.

2. Study context

2.1 Track circuit principle

The track circuit has a central position in the French signalling system: it enables to detect automatically
and continuously the presence of a vehicle on each railway track portion. To do that, a transmitter sends
a current through the 2 rails to a receiver located one or two kilometers away. When a vehicle is on a
track circuit, its first axle achieves a short-circuit, and the signal level at the receiver is breaking down. A
relay, connected between the rails at the end of the track portion, is controlled by this signal. Its
decreasing below a threshold indicates that the track section is occupied. Moreover, in the French high
speed railway lines, the lateral signalling is replaced with a track / vehicle transmission system that uses
rails to continuously transmit coded data to the train, over a specific carrier frequency. This system, called
TVM, provides the driver with essential security informations, such as the maximum authorized speed on
a given section under safety constraints. When the level of the carrier frequency becomes too low, the
train automatically stops. The TVM system uses the same structure as the track circuit: the coded data
are transmitted through the rails as a modulated current that is injected on particular points of the line
named transmission joints. In front of the leading axle of the train, induction sensors capture the
information and transmit it to a receiver, which permits to display a clear text to the driver.
Here we study more specifically one type of track circuit used on high speed railway lines ( UM71C
TVM430). There are 4 types of such track circuits, each with its own carrier frequency. Two frequencies
are allocated to the up line and the other two to the down line. Each track circuit of type UM71C TVM430
consists of (see Figure 1):
- a transmitter that supplies an alternative current at a specific frequency;
- a transmission line constituted by the rails;
- a receiver that achieves a trap circuit to avoid the transmission of information to the
neighbouring section;
- trimming capacitors connected between the two rails at constant spacing to compensate the
inductive behaviour of the track and to maintain the level of the carrier frequency.

track circuit

transmitter Icc receiver

rail

C C Icc C

rail
C : trimming capacitors axle
running direction
Figure 1: Track circuit representation

All these components are subject to dysfunction that should be detected as soon as possible in order to
keep the whole TVM system safe and available. The dysfunction may be due to ageing of the
components, or to track maintenance operations.

2.2 Track circuit maintenance

An inspection car equipped with induction sensors placed in front of the leading axle provides back-up to
the ground-based checks by dynamically monitoring the reception level or the TVM signal picked up by
the moving vehicle throughout the track. The effective current, denoted Icc, is then recorded as a graph
simultaneously for the 4 frequencies of the track circuits used by the TVM system (see Figure 2).

Figure 2: Paper graph recorded by the inspection vehicle

Two kinds of recording are obtained: a paper one, and a digital one. Until now, only the paper version of
the recording is exploited by the maintenance experts. Their diagnosis method mainly consists in
comparing the signals levels to thresholds so as to ensure that:
- the system operates properly in terms of reception level
 - the system does not pick up a signal from an adjacent track transmitter, or a neighbouring
track circuit (electrical separation joint leakage)
Heuristic interpretation of typical parameters (such as the envelope of the signal), also permits to detect
some removed capacitors.
The main drawback of this method is that it is time-consuming and fastidious. On top of that, it only
enables a corrective maintenance, since it is impossible to detect growing defects or to assess the
graveness of the detected defects.
In order to automate the diagnosis task, a pattern recognition method was developed to detect and
localize removed capacitor s. The interested reader should refer to [1], [2], [3]. Here, we only recall the
main steps of this diagnosis method, which are:
- choosing a set of relevant parameters to describe the Icc signals (data representation space).
These parameters have to be enough significant to distinguish between defective track
circuits and standard ones,
- building a classifier (using a neural network for example). The classifier distinguishes
between 2 classes of signal: class 0, corresponding to the “no defect” case, and class 1,
corresponding to the “one removed trimming capacitor” case. The tuning of the classifier
parameters is set using a training database made up of simulated I cc currents.
- In case of “class 1” labelling, localizing the defect, by simulating the I cc signal N times (N is
the number of trimming capacitors along the track circuit). One trimming capacitor is omitted
each time. The simulation that leads to a minimum quadratic error between the experimental
signal and the simulated ones corresponds to the right removed capacitor.
This method gives reliable result s (97.6% of good detection and 0.4% of false alarms), but it permits to
detect only one defect, and this defect can only be a major defect (removal capacitor).
In this work, the aim is to detect all the defects of the track circuit, ranging from a weakly defective
capacitor (with a small resistive behaviour) to a removed capacitor. At the same time, we try to give an
estimation of the graveness of the defects detected, so as to help the maintainers to establish a relevant
schedule of maintenance operations.

3. Diagnosis method based on an electrical modelling

The first step of the automatic diagnosis presented in the previous section was the modelling of the track
circuit [4] so as to simulate I cc signals. This model is interesting at 2 levels: first, it allows to have a better
knowledge about the influence of each component on the Icc signal. Then, thanks to the model, we are
able to build a large database of defective Icc signals, on which advanced signal processing methods can
be applied. The model considers the track circuit as a succession of symmetric and linear quadripoles.
Each quadripole is equivalent either to a homogenous track portion or to a single element (trimming
capacitor, transmitter,...). Moreover, the railway track is considered as a distributed constant transmission
line. For each position of the vehicle, the quadripoles beyond and after the first axle of the vehicle are
estimated, and the Icc current is computed from a set of matricial equations. This model leads to a good
agreement between the simulated signals and the real signals measured by the maintenance vehicle (see
Figure 3).
Case with no defect Case with one defect
4 3

Registered Icc Registered Icc

Simulated Icc Simulated Icc
2.5
3

1.5

1
1
defect

0 0.5
0 250 500 1000 1250 1500 0 250 500 1000 1250 1500

Figure 3: Comparison between real and simulated signals

The electrical model can be used to make the diagnosis of the track circuit. In this case, the idea is to
optimize each parameter of the electrical model, so that the simulated Icc signal matches “exactly” the
signal recorded by the inspection vehicle. Then, we determine which component of the track circuit is
defective by comparing the optimized parameters to their nominal values. In this way, an estimation of the
defect graveness is given by the optimized physical parameters of the track circuit component. The
problem is that the electrical model requires the adjustement of more than 100 parameters (linked to each
component such as the trimming capacitors, the rails…). So it is difficult to optimize all of them in one step
only.
Hence, we intend to use a local approach. Rather than working with the global electrical model presented
in [1], smaller physical models of track circuit portions are implemented, and local diagnosis tasks are
achieved for each of them.

3.1 Local approach of the diagnosis

If we consider that the first axle of the vehicle realizes a perfect short-circuit, the I cc signal at a position P
is totally independent from all the elements of the track circuit located downstream from this position.
Hence, the shape of the signal at point P only depends on the elements located between the transmitter
and the position P. Moreover, when a defect occurs somewhere in the track circuit, the signal is only
modified from the defect to the receiver, while it is unchanged upstream from the defect (see Figure 4).

No defect
Defective capacitor
defect

0 200 400 600 800 1000 1200 1400 1600

Figure 4: Influence of a defect on the Icc signal

Due to the structure of the UM71C TVM430, the choosen local model corresponds to one trimming
capacitor attached with the track portion between this capacitor and the next one. This local model is
called trimming cell. (see Figure 5). All the elements between the transmitter and the cell are seen as an
equivalent Thevenin generator, and all the elements located downstream from it (going towards the
receiver) will be neglected since they have no influence on the Icc signal at the current position.

Cell 1 Cell N

Eeq r1 r2 rN
R
Zeq C1 C2 CN

Transmitter Rail impedance Receiver

Figure 5: Local electrical modelling

Contrary to the global electrical model, which requires about 100 parameters, each trimming cell needs
only 4 parameters, that are:
 - the capacitor value C
- the capacitor additional resistance, r, that models the ageing of the capacitor
- the characteristic impedance of the track Zc
- the propagation coefficient of the track θ [3]
We can note that the defect capacitor is modelled by an additional resistance r, ranging from r = 0 Ω
(perfect capacitor) to r = ∞ Ω (removed capacitor).

3.2 Iterative diagnosis method

The principle of the diagnosis method consists in moving the local model along the track circuit from the
transmitter to the receiver. For each trimming cell, we simulate locally the Icc signal by using quadripole
transfer matrices and the distributed constant line theory. Then, the 4 physical parameters of the current
cell are optimized so that the simulated signal is as close as possible to the measured one. Once it is
done, we study the following trimming cell in the same way, until the whole track circuit is analyzed. A cell
is classified as defective when its optimized parameters differ from the nominal ones.
This iterative method requires a 3-step processing at each iteration :
- determination of the Thevenin generator equivalent to the elements located upstream from the
current cell where the axle is located. The elements downstream from the axle are neglected
under the hypothesis of a perfect short-circuit.
- simulation of the Icc current within the cell where the axle is
- optimization of the 4 parameters by minimizing the mean square error between the simulated and
the experimental signals
Figure 6 illustrates the principle of the diagnosis method.

Figure 6: Principle of the diagnosis method

3.3 Optimization of the physical parameters

To optimize the physical parameters of a trimming cell, we minimize the quadratic error between the
simulated signal (noted Isimu), and the experimental recorded signal corresponding to the same cell (Imeas),
defined as follows [5]:
 J c ( ieC cl, mrl , θeZ a,∫ )s (I s (=i) xIm ( u, Cc− , rx , Z ,θ )) d x
xi 2
(1)
x i−1

(Ĉ, r̂, Ẑ , θ̂ )
c
cell i
= min J cell i (C, r, Z c , θ ) ,
C,r, Z c ,θ
(2)

where interval [x i-1,x i] corresponds to the current cell i.

In the following, we consider that within a cell, only r can vary. C, Zc and θ are fixed, so in the optimization
step, we only adjust the value of the additional resistance.

4. Application to the diagnosis of trimming capacitors

4.1 Results on simulated database

To assess our diagnosis method, a simulated database of Icc signals is computed. This database is made
up of 1000 signals, simulated with the global electrical model of the track circuit. A noise is added to the
signals (using an appropriate model), so that they are as close as possible to experimental r ecorded
signals. The simulations are performed by using several values of characteristic parameters of the railway
track (ballast resistance, characteristic impedance of the track). The number of trimming capacitors of the
track circuit is 19. Among these, 3 cases are possible :
- no capacitor is defective (100 signals)
- 1 capacitor is defective (450 signals)
- 2 capacitors are defective (450 signals)
The position of the defects and the corresponding values of their additional resistances are randomly
chosen, covering all the possible positions, with more or less important defects. For each signal, we apply
our diagnosis method. A capacitor is assigned to the defective class if its additional resistance is greater
than a threshold rth. So we can evaluate the method relatively to the graveness of the considered defects.
Table 1 shows how the results are classified, and table 2 summarizes the results for different threshold rth,
corresponding to 3 levels of defect graveness: not important (ri ≤ 2 Ω), important (2 Ω ≤ ri ≤ 4 Ω), and
critical (ri ≥ 4 Ω).

Reality
Defect No defect
Diagnosis
Good detection False alarm
Defect
(GD) (FA)

No detection Good rejection

No defect
(ND) (GR)

Table 1: Classification of the results

Good detection Good rejection False alarm Non detection

rth=0.9Ω 85.8 % 98.8 % 1.2 % 14.2 %

rth=2Ω 96.1 % 98.4 % 1.6 % 3.9 %

rth=4Ω 97.4 % 98.8 % 1.2 % 2.6 %

Table 2: Results obtained on the simulated database

We can notice that important and critical defects (2 Ω ≤ ri ≤ 4 Ω and ri ≥ 4 Ω) are very well detected (96.1
% and 97.4 % of good detection). Moreover, we have quite few false alarms (1.2 %), which is also
interesting. Nevertheless, when the defects are less important (ri < 2 Ω), the results are less reliable,
which is highlighted by a high number of non detections (14.2 %). Further works will focus on this point.
 4.2 Some results on real signals

In complement with this evaluation on simulated database, we have applied our diagnosis method on real
data. 2 examples of defective track circuits are presented. These 2 track circuits are completely identified
by electrical measurement of the trimming capacitors impedances. For the first one, only capacitor #9 is
defective (r9 = 3.07 Ω). For the other one, only capacitor #23 has an additional resistance r23 =
1.74 Ω. The results of the diagnosis are summarized in Table 3. We note k the position of the defect for
each track circuit.

rˆi estimated ri measured

rˆk estimated rk measured
i=1,…,N , i≠k i=1,…,N , i≠k
Track circuit n° 1
k=9
rˆi < 0.3 Ω ri < 0.2 Ω r̂9 = 11.55 Ω r9 = 3.07 Ω

Track circuit n° 2
k=23
rˆi < 0.7 Ω ri < 0.1 Ω r̂23 = 1.77 Ω r23 = 1.74 Ω

Table 3: Application of the diagnosis method on 2 real signals

We can notice that each defect is well detected and localized but the estimated value of the resistance is
not very accurate. In fact, in order to work well, this method needs an accurate tuning of the railway track
parameters such as ballast resistance or characteristic impedance. A lack of knowledge of those
parameters leads to an error in the internal resistance estimation. However, even if the estimated value is
not the real one, it can be considered as a good indication of the defect graveness (important defect for
the first track circuit, not important for the second one).

5. Conclusion

We have proposed a diagnosis method based on a local modeling of the track circuit to detect defective
trimming capacitors of the TVM signalling system. This diagnosis is quite efficient, since it allows to detect
all the defective capacitors of a track circuit, while giving an indication of their level of graveness. The
results on both simulated and real signals are very encouraging. The defect localization is quite good, but,
the graveness estimation has to be improved. This is due to a bad estimation of some electrical
parameters of the track. To increase the diagnosis efficiency, a mixing approach will be developed, that
associates the physical parameter optimization with a black-box model using the fusion theory [6].

Acknowledgements

This work is supported by the Infrastructure Department and the Engineering Department of the French
Railway Company (SNCF).

References

[1] P. Aknin, L. Oukhellou, F. Vilette. Track circuit diagnosis based on automatic analysis of measurement
records, in WCRR 2003, Edinburgh
[2] L. Oukhellou, P. Aknin, A. Debiolles, F. Vilette. Automatic diagnosis of track circuit in predictive
maintenance context, Railway Engineering, 2003
[3] L. Oukhellou, P. Aknin, F. Vilette. Classification and localization of defects by combined use of internal
and external modelings. PSIP, Grenoble, France, 2003
[4] F. Fessant, P. Aknin, F. Vilette, M. Antoni. Modélisation électrique du circuit de voie, élément du
système de transmission voie-machine des TGV. 3EI revue, n°27, 2001
[5] J.A. Nelder, R. Mead. A simplex method for function minimization, Computer journal, vol.7, p.308-313,
1965
[6] A. Debiolles, T. Denoeux, L. Oukhellou, P. Aknin. Evidential fusion fo classifiers relating to spatially
dependent subsystems. Application to a railway device diagnosis. IPMU 2006, Paris, July 2006

View publication stats