![[Translate to English:] Frontalansicht ICE-Zug mit DB-Logo.](/fileadmin/_processed_/6/6/csm_DB190480_937ae4df28.jpg)
In rail, the possibilities to gather data and use Data Mining methods have so far been limited: oftentimes, operational data are collected, concentrated and interpreted; however, systems such as ICEs only allow a manual collection of data, e.g. via USB stick from a black box, the so-called Juridical Recorder. Therefore, it is necessary to investigate how data can be automatically collected, processed and interpreted in order to generate the highest possible added value.
In complex systems such as the ICE, only a few errors can be traced back to individual components - the interaction of several components as well as the resulting behaviour plays a key role. The resulting data are very heterogeneous and usually depend on the respective operating status.
In the event of operational disruptions or unexpected misbehaviour of systems, a large number of experts and a great deal of time are often required to localise the causes of errors. One reason for this is the large number of recorded data time series that come from heterogeneous sources, ranging from sensor data to operational data telegrams to human-machine interactions.
A purely machine-based or a machine-human-assisted evaluation of these data can provide a much more efficient clarification of malfunctions and also enable predictable maintenance. However, the evaluations and the associated actions should be self-explanatory and comprehensible for experts.
"We have investigated the extent to which predictive maintenance can be carried out on faults within an ICE system using the example of air conditioning," explains Dr Sven Schmeier, Chief Engineer at the DFKI in Berlin. "In addition to classifying 'There will be a fault in the next few days', the corresponding reasoning is also provided, for example 'The outside temperatures are above 30 degrees Celsius, the fan centrifuge shows vibration tolerances and the cooling pipe cleaning was two months ago'."
The AGEME project described a concept for an explainable, data-driven warning system for complex systems such as an ICE. The concept provides for a standard data interface and thus also makes a significant contribution to the standardisation roadmap AI of the Deutsches Institut für Normung e.V. (German Institute for Standardisation). (DIN), the German Commission for Electrical, Electronic & Information Technologies iand the Federal Ministry for Economic Affairs and Energy (BMWi). It also decouples the application from the AI analysis, so that AI procedures also become easier to apply and more comparable.
The explanations for the warnings are presented in a form that is useful for the experts involved on the one hand, but also for users who are not familiar with AI. Whether and how such a warning system can be effectively and fully implemented from a scientific point of view and whether the results can lead to further automation from a railway point of view is to be researched in a follow-up project.
AGEME was funded by the Federal Ministry of Transport and Digital Infrastructure (BMVI) under the Modernity Fund ("mFUND") funding line with €98,477.60.
In rail, the possibilities to gather data and use Data Mining methods have so far been limited: oftentimes, operational data are collected, concentrated and interpreted; however, systems such as ICEs only allow a manual collection of data, e.g. via USB stick from a black box, the so-called Juridical Recorder. Therefore, it is necessary to investigate how data can be automatically collected, processed and interpreted in order to generate the highest possible added value.
In complex systems such as the ICE, only a few errors can be traced back to individual components - the interaction of several components as well as the resulting behaviour plays a key role. The resulting data are very heterogeneous and usually depend on the respective operating status.
In the event of operational disruptions or unexpected misbehaviour of systems, a large number of experts and a great deal of time are often required to localise the causes of errors. One reason for this is the large number of recorded data time series that come from heterogeneous sources, ranging from sensor data to operational data telegrams to human-machine interactions.
A purely machine-based or a machine-human-assisted evaluation of these data can provide a much more efficient clarification of malfunctions and also enable predictable maintenance. However, the evaluations and the associated actions should be self-explanatory and comprehensible for experts.
"We have investigated the extent to which predictive maintenance can be carried out on faults within an ICE system using the example of air conditioning," explains Dr Sven Schmeier, Chief Engineer at the DFKI in Berlin. "In addition to classifying 'There will be a fault in the next few days', the corresponding reasoning is also provided, for example 'The outside temperatures are above 30 degrees Celsius, the fan centrifuge shows vibration tolerances and the cooling pipe cleaning was two months ago'."
The AGEME project described a concept for an explainable, data-driven warning system for complex systems such as an ICE. The concept provides for a standard data interface and thus also makes a significant contribution to the standardisation roadmap AI of the Deutsches Institut für Normung e.V. (German Institute for Standardisation). (DIN), the German Commission for Electrical, Electronic & Information Technologies iand the Federal Ministry for Economic Affairs and Energy (BMWi). It also decouples the application from the AI analysis, so that AI procedures also become easier to apply and more comparable.
The explanations for the warnings are presented in a way that is useful for the experts involved, but also for users who are not familiar with AI. Whether and how such a warning system can be effectively and fully implemented from a scientific point of view and whether the results can lead to further automation from a railway point of view is to be researched in a follow-up project.
AGEME was funded by the Federal Ministry of Transport and Digital Infrastructure (BMVI) under the Modernity Fund ("mFUND") funding line with €98,477.60.