Sie können Operatoren mit Ihrer Suchanfrage kombinieren, um diese noch präziser einzugrenzen. Klicken Sie auf den Suchoperator, um eine Erklärung seiner Funktionsweise anzuzeigen.
Findet Dokumente, in denen beide Begriffe in beliebiger Reihenfolge innerhalb von maximal n Worten zueinander stehen. Empfehlung: Wählen Sie zwischen 15 und 30 als maximale Wortanzahl (z.B. NEAR(hybrid, antrieb, 20)).
Findet Dokumente, in denen der Begriff in Wortvarianten vorkommt, wobei diese VOR, HINTER oder VOR und HINTER dem Suchbegriff anschließen können (z.B., leichtbau*, *leichtbau, *leichtbau*).
Dieses Kapitel befasst sich mit der Anwendung der Distributed Acoustic Sensing (DAS) -Technologie zur Erkennung von Defekten an Schienen, mit besonderem Schwerpunkt auf Schienenverbindungen. Die Studie, die in Zusammenarbeit mit der Société du Grand Paris durchgeführt wurde, untersucht, wie DAS Eisenbahnschienen auf Defekte überwachen kann, bevor sie zu Sicherheitsproblemen werden. Der vorgeschlagene Ansatz beinhaltet die Automatisierung des Erkennungsprozesses mithilfe maschineller Lerntechniken, insbesondere Support Vector Machine (SVM) und XGBoost-Klassifikatoren. Die Studie analysiert die Unterschiede in den Schwingungsmustern zwischen zwei Arten von Glasfaserkabeln, einem verlegten und einem nicht verlegten, und extrahiert charakteristische Merkmale im Zusammenhang mit Schienenverbindungen. Die Ergebnisse zeigen eine hohe Genauigkeit bei der Erkennung von Schienenverbindungen, wobei der XGBoost-Klassifikator einen F1-Wert von 0,93 erreicht. Die Studie diskutiert auch das Potenzial zur Erkennung anderer Defekte an Schienen und die Bedeutung der Einbeziehung von Feldinformationen, um die Zuverlässigkeit der Erkennung zu erhöhen. Dieses Kapitel bietet einen umfassenden Überblick über die Effektivität der DAS-Technologie bei der Verbesserung der Eisenbahnsicherheit und der Automatisierung von Fehlererkennungsprozessen.
KI-Generiert
Diese Zusammenfassung des Fachinhalts wurde mit Hilfe von KI generiert.
Abstract
In recent years, advances in Distributed Acoustic Sensing (DAS) technology have resulted in significant progress in the detection of vibration sources. However, its use in railway monitoring is still relatively new, even though thousands of kilometers of optical fiber cables are already set up for telecommunication purposes, thus potentially exploitable.
In this paper, we explore the possibility of using a DAS system and machine learning tools to detect rail defects along the track. Rail defects are defined as anything other than a smooth rail, and we focus on the detection of rail joints, which are common elements along the track. In this study, measurements were carried out on a short railway section of a few kilometers between two train stations in Paris. The results show that nearly all rail joints along the track are correctly detected, demonstrating the ability of the system to detect these elements with a spatial accuracy of a few meters. Lastly, some future perspectives for the study are proposed, such as a more in-depth analysis of the detected locations or the integration of field information to enhance the reliability of detections.
1 Introduction
Using optical fibers for data transmission has transformed the way we communicate, enabling instant transfer of information over long distances. Yet, despite its potential, little work has been done in the railway industry. Recent works have shown that DAS technology can detect a wide range of vibration events in the railway environment, such as a moving train, or rock falls. Specifically, when a train passes over a rail, it produces an acoustic vibration that could contain valuable information about the track condition.
For example, apart from a broken rail, if rail defects are not quickly identified and repaired, they can ultimately cause a train to derail. As such, they represent an important issue for rail safety. Currently, most defects are detected either by inspection vehicles, visual inspection, signaling systems or derailment [1].
Anzeige
Unlike these methods, DAS technology does not require an expensive deployment on the tracks. All that is required is a connection to an optical fiber and a single device can monitor a line up to 193 km [2]. This method can therefore be used for continuous track monitoring to detect defects before they become safety problems. Several studies have already analyzed the ability of DAS to monitor the track [3] and detect broken rails [4]. An in-house study has also been carried out demonstrating the feasibility of broken rail detection using DAS and signal processing techniques [5].
This paper proposes to automate the detection process using machine learning techniques, which has the advantage of being less dependent on parameters. We focus on the detection of rail joints as they are common elements along the track and exhibit behavior similar to rail cuts [5]. This article is the result of a collaboration with Société du Grand Paris (SGP) in charge of the Grand Paris Express program. Covering nearly 200 km, this major urban program seeks to connect suburban areas without passing through Paris, by developing new metro lines and extending existing ones.
2 Distributed Acoustic Sensing
Distributed Acoustic Sensing (DAS) relies on Rayleigh backscattering. When light travels through the core of an optical fiber, some of it is scattered due to interactions with the fiber material imperfections. A portion of this scattered light is reflected to the source. Any vibration source near the optical fiber will affect both the amplitude and phase of this signal. These vibrations can hence be detected and monitored by recording and analyzing the backscattered light.
In this study, the DAS system uses coherent heterodyne detection [5], with an optical frequency shift of 18 MHz. The acquisition is performed at a sampling rate of 125 MHz. The pulse rate is 4 kHz with a spatial resolution of 10 m.
Anzeige
3 Proposed Approach
The study was carried out in an environment similar to the new metro lines under construction by SGP, i.e. a few kilometer railway section between two stations in Paris connected by an underground tunnel. The study section is equipped with numerous fibers with different layouts resulting in an optical fiber length of approximately 20 km.
Two types of fibers run along the tracks: one is placed under the concrete beneath the tracks (type 1), and the other is laid in the cable trays attached to the tunnel structure (type 2).
The study section has been configured to ensure coverage of both tracks in the test site and to alternate between the use of the two types of fibers.
3.1 DAS Data
For our purposes, we use the differential phase computed from the DAS phase data. It is then pre-processed by applying a high-pass filter, followed by a normalization step based on the estimated noise standard deviation. The result is a two-dimensional waterfall of the vibration activity over time along the track. Figure 1 shows an example of a waterfall.
Fig. 1.
A 5-min waterfall for two optical fiber cables in the study site showing vibrations induced by moving trains.
Reproduced with permission from SNCF Réseau, copyright SNCF Réseau, 2024
Given the large number of optical fiber cables available, the study will focus on only two cables, one of each type: cable 2 (type 2) and cable 6 (type 1).
Figure 1 shows the vibrations activity when train wheels pass over the tracks. When they encounter a rail joint, a specific pattern can be observed. Figure 2 shows several examples of rail joint vibrations on different rail joints along the tracks.
Fig. 2.
Rail joint vibrations on different rail joints along the tracks. Data from the first row comes from cable 2, while the second row comes from cable 6.
Reproduced with permission from SNCF Réseau, copyright SNCF Réseau, 2024
Rail joint vibrations are more intense and regular over time. These high vibration intensities correspond to the impact of all the train's wheels on the rail joint.
In Fig. 3, vibrations measured from optical fiber cable 2 show a track section of approximately 500 m containing two rail joints. The distinction between the vibrations generated by a track in good condition and a joint is clearly observable. Similar observations can be made when examining data from cable 6.
Fig. 3.
Zoom on rail joint vibrations observed as a train travels along a track section (first row: data from cable 2, second row: data from cable 6).
Reproduced with permission from SNCF Réseau, copyright SNCF Réseau, 2024
We can notice that vibrations between the two fiber cables are quite different. These differences can be explained by the way the cables are installed. Data from cable 2 shows more intense vibrations and more noise caused by interferences with surrounding vibrations, as this cable is not buried and therefore more sensitive. In contrast, vibrations from cable 6 are less intense (buried cable), less noisy, but sharper and clearer.
By considering the features mentioned above, a rail joint detection algorithm can be proposed.
3.2 Detection Algorithm
The proposed detection method consists of two steps. First, the signal of interest, i.e. the one associated with the moving train, is detected. Next, the rail joint positions are identified using a classification model trained to classify the signals into two classes (Track in good condition / Rail joint). To ensure an accurate classification, a set of distinctive features related to joints must be extracted and used.
Several studies have been conducted and the most relevant features found include energy, autocorrelation and a frequency analysis of the frequencies excited by the joints.
The importance of the energy feature is quite apparent based on the vibrations observed in Figs. 2 and 3.
The periodic aspect of these vibrations was highlighted through an analysis of the autocorrelation signal. Autocorrelation consists in quantifying the similarity between a signal and a time-shifted version of itself. It is therefore well-suited for discerning patterns in a signal.
Additionally, a frequency analysis showed that the rail joint information fell within a particular frequency range. Figure 4 shows the distribution of frequency peaks in the frequency signal from cable 2. Green vertical lines correspond to identified rail joints positions in the field. We can notice that most peaks are located at rail joint positions. However, other locations also show a strong frequency content. As the selected features are weakly correlated between each other, these positions will be later excluded through the combined use of features by the classifier.
Fig. 4.
Distribution of frequency peaks in the signal from cable 2 (range [4.5 Hz – 8.0 Hz]).
After features were extracted, a classification model was trained using Grid Search with k-fold cross validation to find the best parameters. Several classifiers were studied including Support Vector Machine (SVM) and XGBoost.
Table 1 shows the performance of our algorithm, obtained on a dataset of approximately 15,000 samples. For the calculation of the evaluation metrics presented below, around 4000 samples were used.
Table 1.
Performance of the proposed rail joint detection algorithm
Accuracy
Precision
Recall
F1-score
SVM
0.906
0.91
0.90
0.91
XGBoost
0.927
0.93
0.93
0.93
It should also be noted that the ground truth is reliable within a 5-m margin which has an impact on the performance of the algorithm. It only considers positions of \(\pm\) 5 m around the ground truth as joints. Therefore, slightly off-center joints outside of this scope are considered as false positives, thus lowering the algorithm accuracy.
4.2 Results Visualization
Figures 5 and 6 show results obtained on optic fiber cable 2 respectively on a 5 min and 25 min of peak hour traffic between 6PM and 7PM. We can see that nearly all rail joints have been detected several times (except for example at ~ 700 m and ~ 3200 m). This shows that the algorithm is consistent and repeatable. The solution can detect the same joint on multiple occasions. And when it does not detect a joint (positions listed above), these positions never seem to be detected later.
In addition, many railway equipment is located at the entrance and exit of train stations, which explains the high number of false detections at these locations.
Similar results have been obtained from cable 6 data, using the same approach. Thus, the proposed method is effective on both types of optical fibers. However, only features extracted from the direct measured vibrations have been exploited. Further research will involve adding field information about the track section (its geometry, axle loads or rolling stock types) that can strengthen the reliability of detections.
Fig. 5.
Detected rail joints on 5 min of peak hour traffic (XGBoost, cable 2).
Reproduced with permission from SNCF Réseau, copyright SNCF Réseau, 2024
This study shows the effectiveness of using Distributed Acoustic Sensing technology to detect rail joints. These elements leave a discontinuity between two rail segments in a railway track which generate an acoustic vibration that is different from normal train vibrations. We can assume that other rail defects can be similarly detected, as they also have an impact on the track by breaking its smooth surface. The challenge is then to be able to separate and identify these defects from each other.
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
Kumar, S.: Study of Rail Breaks: Associated Risks and Maintenance Strategies, Technical Report. Lulea University of Technology, Lulea, Sweden (2006)
2.
Ayad, T., Kabalan, A., Garbini, G.P., Hammi, T.: Over 193 km sensing range by optical fiber distributed sensor with mixed amplification repeaters for railway field applications. Opt. Sensors (2023)
3.
Cedilnik, G., Hunt, R., Lees, G.: Advances in train and rail monitoring with DAS. Opt. Fiber Sensors, ThE35 (2018)
4.
Wagner, A., Nash, A., Michelberger, F., Grossberger, H.: The effectiveness of distributed acoustic sensing (DAS) for broken rail detection. Energies 16(1), 522 (2023)
5.
Ruffel, M., Ben Amara, I., Hammi, T., Kabalan, A.: Broken rail detection using distributed optical fiber sensing technology. In: World Congress on Railway Research 2022, Birmingham, United Kingdom (2022)
