Data analysis and predictive maintenance for rail vehicles
Every year, Deutsche Bahn transports well over 100 million passengers in national and international long-distance traffic. With comprehensive digitalization, it is making itself fit for the future.
Together with optiMEAS DB Systemtechnik GmbH is developing a predictive maintenance system with the aim of predicting the "state of health" of relevant drive components of ICE's and being able to schedule maintenance work in depots and workshops in a targeted manner.
In order to develop predictive maintenance algorithms, mathematical-physical models that describe the regular behavior are required first. In order to be able to compare these models with the real conditions, the operating data of the components must be recorded.
The measurement data is acquired at high sampling rates and stored in the central cloud (optiCLOUD, called FALKOS in the DB) as historical data for analyses.
Every day, around 50 gigabytes of data are collected. To be able to calculate this amount of "big data" according to the models, DB uses an analysis cluster based on Hadoop for parallel processing. Current methods for machine and deep learning and parameter determination for neural networks are used. Technologically, open source frameworks such as Tensorflow or Keras are behind it.
To be able to apply artificial intelligence in practice, perhaps the most important component is the domain knowledge from the specialist departments and workshops. This is also the experience of DB Systemtechnik's engineers: it is only by combining mathematics, IT and application knowledge that patterns can be identified that allow anomalies to be detected and recommendations made for maintenance.
Through the cooperation with optiMEAS we have already learned a great deal about the vehicles, which can be used for maintenance planning and improves the quality and availability of the fleet.