OnboardEU – automatically detect damage to the track with AI
OnboardEU – automatically detect damage to the track with AI
With the OnboardEU project, the DLR has set itself the goal of developing quieter and more efficient trams. The automatic detection of damage to the rails and track superstructure using suitable AI processes and noise mapping for the targeted implementation of countermeasures are intended to help achieve this.
Trams are highly frequented means of public transportation that also run through densely populated city districts. Passengers and residents usually find the noise and vibrations caused by damage to the tracks particularly annoying. The damage also means higher costs for the vehicle operators. This is where the OnboardEU project comes in, in which the German Aerospace Center (DLR), together with the AIT Austrian Institute of Technology GmbH, Vienna, and i4M technologies GmbH, Aachen, is developing solutions for the automatic detection of damage using AI. The project is being funded by the Federal Ministry for Digital and Transport (BMDV) as part of the mFUND innovation initiative with around 760,000 euros over a period of three years until the end of 2024.
"The aim of the project is to research suitable AI methods for the automatic detection of damage to the rails and track superstructure. To this end, we will continuously record the dynamic vehicle reactions during operation and evaluate them on the vehicle using AI methods," explains project coordinator Dr. Jörn Groos from the DLR-Institute of Transportation Systems. "We are also working on noise mapping. This will allow us to identify stretches of road with particularly high noise emissions so that we can take targeted countermeasures."
For OnboardEU, the researchers are equipping twelve vehicles in various European cities (including Hanover and Düsseldorf) with onboard measurement systems with edge computing capability. These systems record signals from acceleration sensors (vibrations/shocks) and microphones (direct sound/noise) in the area of wheel-rail contact. The measurement data collected can be localized precisely on the track using a map-based linking of several sensors. AI algorithms are developed and tested on the basis of the collected vibro-acoustic data and additional surveys of the track condition (e.g. recognizable damage to the rails such as corrugations). A key challenge is the development of robust automatic evaluation methods for the very complex and demanding measurement data. In the future, fully automatic evaluation should enable low-cost, low-effort condition monitoring during operation.
The aim is to support the development of corresponding AI processes beyond the project and the project participants. Another goal of OnboardEU is therefore to create a training data catalog that is open to all stakeholders. This will be available via the mCLOUD after completion of the project in May 2025. "Diverse and comprehensive data sets are a prerequisite for the successful training of reliable and robust AI approaches. The availability of corresponding open data sets promotes the ready-to-use implementation and thus broad application of corresponding approaches in practice," emphasizes Groos.
OnboardEU – Detection of track defects and creation of immission maps in European cities based on vibration and acoustic data measured in regular traffic. The collected measurement data is processed, georeferenced and analyzed on a background system. Conspicuous vehicle reactions and increased noise emissions in the rail network can then be efficiently mapped.
Credit:
AIT
i4M technologies GmbH develops and manufactures the onboard measuring systems, including the associated sensors. The focus here is on designing the measurement systems in such a way that AI algorithms can analyze the incoming sensor data on the train in real time. This enables a significant reduction in data from big data to smart data.
In the project, the Transportation Infrastructure Technologies department at AIT is taking on the task of supervised machine learning to automatically detect track defects based on sound and vibration data. At the same time, the sound propagation of different vehicles and operating conditions is to be investigated by means of pass-by measurements in order to support the mapping of noise caused by trams .
In the project, the DLR Institute of Transportation Systems is responsible for implementing track-accurate georeferencing, including making it available as open source software. The scientists are also developing machine learning methods, particularly in the area of unsupervised learning, and preparing the training data set.
About the mFUND funding program of the BMDV
As part of the mFUND funding program, the BMDV has been supporting research and development projects relating to data-based digital innovations for Mobility 4.0 since 2016. The project funding is supplemented by active professional networking between stakeholders from politics, business, administration and research and the provision of open data on the mCLOUD portal. Further information can be found at www.mfund.de.
