Good vibrations – Distributed Acoustic Sensing for railroad tracks

The Device Chronicle interviewed Sebastian Haid, Product Owner, Train Tracking and Information Provisioning, Frauscher Sensonic to learn about how Distributed Acoustic Sensing (DAS) is advancing the sensing of potential adverse impacts on railway tracks. 

Sebastian is trained as an astrophysicist and graduated with expertise in magneto-hydrodynamics, big data analysis, high-performance computing, simulations and theoretical physics. Sensonic is part of the Frauscher group and develops railway specific solutions based on DAS. DAS is an excellent technology for sensing vibrations in the vicinity of fiber optic cables and tracks along which these cables are laid. These vibrations can then be quantitatively analysed to identify moving objects, people and events along railroad tracks.

Distributed acoustic sensing expert Sebastian Haid
Distributed acoustic sensing expert Sebastian Haid

Distributed acoustic sensing comes to rail

Sebastian begins by explaining that train tracking information provisioning is a growing field. Distributed Acoustic Sensing has been used in many industries since the 1970s for application in oil and gas such as pipeline monitoring and perimeter defence. Its use in rail came later in 2014 when Frauscher – the parent company of Sensonic – introduced a sensor product into the segment. Sensonic has since innovated further and by collaborating with many progressive partners from the sector, Sebastian and the development teams have adapted the sensor technology to make appropriate solutions for important use cases such as condition monitoring. The partners provide access to the base information and then Sensonic performs the data processing and analysis. The partners include freight lines with single tracks and 2 km long trains full of iron ore, regional lines carrying passengers; mainlines with mixed traffic and finally high speed lines with high speed passenger trains. Sebastian explains that Sensonic addresses all the rail segments with the most challenging being tramways. He explains the challenge here: “Laying fibre to do sensing on trams is very difficult as these vehicles share the same road space as heavy trucks and so it is challenging to disambiguate vibrations coming from these two different sources.”

Explaining how distributed acoustic sensing works

Next, Sebastian explains how DAS works. It is, he says, a “80 km long microphone which is one sensor. Whenever or wherever something is happening on or near the track, we can measure the vibrations. And vibrations are powerful representatives of physical processes within mechanical systems. Daily life examples are the stethoscope for medical use or ultrasound for non-destructive workpiece examination. The continuous monitoring of a railway track generates huge amounts of data from which we must extract insights from the data. Sensonic identifies trains, monitors the wheel-rail interface and looks for environmental events. Then from the data, insights can be gathered to answer a series of critical questions such as:

  • Where are trains on the installation
  • On which track are they travelling
  • What is their speed and travelling direction
  • Is there something wrong with the wheel or the track? 
  • Are there people working near the track? 
  • Are there people trespassing over the track
  • Have obstacles dropped to the track (e.g. rocks, trees)?

Sebastian and the development teams track all of this information which is pre-processed with AI. He explains “The train will run from A to B so the data processing for train tracking in this regard is not particularly challenging. It gets more challenging when you look at the wheel-rail interface itself. Each train is a vibrational source that probes the installations and creates a vibrational twin of the installation, respectively of itself. From this so-called SonicTwin real time information about sudden changes such as rail breaks can be extracted. Additionally, SonicTwins are stored which in the end provides an evolutional view of the track degenerating and the substructure changing. From this predictive maintenance is not far away.” Sebastian also says that in advanced rail markets such as central Europe, there is a misnomer that rail breaks do not occur. They are real but identified quickly enough and repaired before something dramatic happens. In other regions of the world this is not the case and fatal accidents occur. The classical way is to send employees along the track to do inspections, use the train driver’s perception or use track circuits. All three methods have significant drawbacks and limitations. “Glass fibres don’t care about experience, or rain or stones falling on it” he concludes.  

Acquiring and processing the data

DAS is the leading edge approach but almost 10 terabytes of data per day is processed on each sensor using edge computing and the newest artificial intelligence algorithms. Sensors run side by side. To train the machine learning models to detect trains and environmental events, labelled data is required. As model training takes time, selected data and ground truth is transferred to the cloud. Sebastian explains that the cloud provides the resources the development teams need.

For the customers the edge devices, which include the optical sensors and a processing unit, convert the vibrational data into a continuous stream of information that is provided to the cloud. The challenge is that the rail operators have some reservations about the cloud. Sebastian says the operators can “think in terms of boxes carrying data packets. This is where the partners are important as they are willing to take the step and connect a sensor to the cloud. Many operators are more conservative, risk averse and are afraid that a third party could enter a sensor and make modifications. But the change is happening as more executives see the benefits of digital adoption. Not only is the information that we provide unique and has a high quality. We continuously improve our models based on the sensor data itself using the expertise of the operator and ground-truth leading to even better quality and more insights.”

Updating the sensor boxes

Sebastian explains that currently all systems have to be physically connected to a network. A VPN connection is used to apply changes to the software on the edge devices. Over-the-air updates are at a very early stage. AWS and Azure are used to store data, train models and from there those are deployed to a repository. Software changes and the newest algorithms can then be brought automatically to the edge devices in the field. In railways, the idea of continuous delivery challenges the mindsets of most of the large railway operators. Usually access to systems is provided singularly and at predefined times. Still, Sensonic drives to deploy as often as needed and as quickly as possible with permanent remote access to monitor and interfere with the devices in the fields continuously. This enables new commercial schemes such as Software-as-a-Service with aim to make the railway market more efficient. “We want to provide the best information service to the customer”. Sebastian says. Sensonic has been thinking about OTA updating of the remote devices: “If you take a remote position, how do you update and operate the sensor systems? We have to get information to the sensors as the number grows. From a product management perspective, installations and updates should primarily be done remotely as each visit at site significantly ties up the resources of the operations team, hence leads to significant costs. The Pandemic has taught us that it is a huge advantage that boxes can be installed and be operated remotely.”

We wish Sebastian and his colleagues well as they advance the use of DAS and AI to serve more use cases in train and track management. 


Recent Articles