We will present our findings on the results of distributed acoustic measurements on optical fibers for the prevention and detection of anomalies along the railway infrastructures. This system can also be a new approach for Structural Health Monitoring (SMH).
We use a commercial Distributed Acoustic Sensor (DAS) unit for these measurements. The DAS unit involves propagating a high frequency light pulse through a standard optical fiber and alayzing the backscattered signal (Ratleigh scattering). Any event occurring in the fiber that has an acoustic energy (vibration for example), will disturb the backscattered signal by shifting its phase . This phase shift is analyzed by the DAS unit in order to provide the nature of the event. The exact location is then obtained by means of reflectometry (OTDR).
We will briefly introduce the results of the acoustic signal processing collected from DAS measurements along two railway infrastructures:
– A portion of high-speed line (40km monitored) which includes several railway viaducts.
– A subway line (2.53km monitored).
For these measurements, we connected the DAS unit on fiber optic cable, which was already installed by the rail operator. It’s a standard telecom optical fiber operated at a wavelength of 1550 nm. One of the advantages of this specific type of measurement is that is uses optical lines which already exist, this greatly reduces the cost of deployment.
Figure 1 : Presentation of the experimental setup
In table 1, the parameters we implemented for each measurement can be seen.
Table 1: Parameters of measurement set ups
Because of the fiber length deployed, we have preferred a low pulse repetition frequency (PRF). Moreover, having a high PRF may cause overlapping pulses and therefore a loss of information.
In this section, we will show the acoustic data measured by the DAS during the passage of the train on the viaduct (see Figure 2). Both air and solid vibrations are digitized. Figure 3 displays the total accumulated spectral energy envelope. In other words, the intensity of the air and solid sounds recorded by the fiber during the passage of the train on the viaduct, and accumulated over the entire recorded frequency band.
Figure 2: Viaduct (480m long) monitored by Cementys
Figure 3: Behavior of the viaduct during the passage of the train and outside the passage of the train
Each point on the viaduct has a specific signature in comparison to the reference signal in blue. If one wishes to associate each signature with an element of the viaduct, they must subtract the reference signal to the signal measured during the passage of the train. By analyzing a 20-meter portion of viaduct located between two beams during the passage of the train, we have plotted in firgure 4 the acoustic spectrum before the passage of the train and after it. These plots exhibit specific excited frequencies.
Figure 4: Acoustic spectrum resulting from the passage of the train on viaduct
The acoustic spectrum corresponds to an average of 70 points of measurements, 1 point for every meter. We can conclude that the frequency f1 (see Figure 4) corresponds to the first flexural mode of the viaduct . Considering a distance between axles of the same bogie of 2.5m, the frequency associated with this speed can be estimated as: 𝑓 =V/D. Being f the frequency (Hz), V vehicle speed (km/h) and d the distance (m). Therefore, the frequency associated to the passage of the axles at 220km/h in a portion of 70m is 3.33Hz. It’s clear from this analysis that maintenance problems of both track and vehicle, situations of passengers discomfort or even lack of security would be derived from the excitation of the first flexural mode of vibration. A more complete dataset on different train passes can lead to further statistical studies and thus to better conclusions. Another analysis on the noise generated by the wheel-rail interaction can lead us to better differentiate the frequencies.
The following study, on a portion of subway line, presents results of the passage of two trains which we can compare. Some track devices have specific signatures as will be shown later:
Figure 5: Comparison between 3 total accumulated spectral energy envelope displays
Figure 6: Correlation between 2 signals resulting from two different passages of train
A first conclusion can be drawn about the nature of the resulting signals on different track devices, a specific intensity to each device. This intensity (see Figure 7) can be monitored by comparing it with a reference signal which results from a train switch in good working order. We follow the evolution of the different signals resulting from different train passages. A decision support system concerning the state of the rail switch can then be put it place. Figure 8 shows the high correlation coefficient between two train passages. In this case, r = 0.78
Figure 7: Representation of the mechanism of generation of rolling noise
Figure 8: Acoustic spectrum during the passage and nonpassage of the train on the rail switch
Another conclusion can be made on the acoustic spectrum generated by the passage of the train on the rail switch. By analyzing these resonance frequencies and associating them with each element of the infrastructure, one can rule that each element is exited at a given frequency. Taking these frquencies as references; 26.36 Hz, 37 Hz, 57 Hz, 92 Hz (which are generated when the operating state of the switch is good) and following the evolution of these frequencies can be an alternative of monitoring this track and rolling device in addition to the intensity. For example, the sleepers vibrate at low frequencies, up to about 500 Hz . Rolling noise covers a large frequency bandwidth, between 100 and 5000 Hz .
The acoustic measurement distributed by optical fiber seems to be a new alternative for the monitoring of railway infrastructures. It’s a low-cost technique and it provides a good prediction of the state of the infrastructure and some track devices. It’s also expected that with a larger data set, such as; the average age of the trains, the weight and the speed of the trains, the state of the rails, several passages of trains (to deepen the statistical measurement) and using the PCA method (principal analysis component), for example, will provide for better predictive maintenance.
 Zhou Sha., et al., Phase demodulation method in phase sensitive OTDR without coherent detection. Opt.
Express, Vol. 25, Issue 5, pp. 4831-4844 (2017)
 Xia, H., Zhang, N., Guo, W.W.: Analysis of resonance mechanism and conditions of train-bridge system.
Journal of Sound and Vibration 294, 810–822 (2006)
 Virginie Delavaud, “Time modeling of wheel / rail interaction, for rolling noise application train” Thesis in
applied physics, p30, (2011)
 D.J. Thompson. Railway noise and vibration: mechanisms, modelling and means of control. Elsevier, 2009