Felix Prause successfully defended his PhD thesis on “Predictive Maintenance in Rolling Stock Rotation Planning” at the Institute of Mathematics at Freie Universität Berlin on February 20, 2026. Indeed, keeping trains in good technical condition is a challenging problem, and methods to improve vehicle health and/or save costs are definitely needed. The availability of sensor data makes it possible for the first time to change from preventive maintenance, which is based on the current condition, to predictive maintenance, which is based on the likely future condition. This promises more for less, i.e., a better vehicle health at simultaneously lower costs. However, it turns out to incur hedging costs, in this case, long deadhead trips, such that railway companies will not use this method easily. Felix Prause has shown that how this problem can be overcome by regular updates of the forecasts using Bayesian inference, such that 10\% of maintenances can be saved as well as 2.5% of costs. The method, which is based on an adaptive discretization of the parameter space of the involved failure distributions, is completely general and carries over to a multitude of further applications.
The research was carried out within the Berlin Digital Rail (BerDiBa) project, which was supported within the funding program ProFIT by the Senate of Berlin’s Department for Economics, Energy and Energy and Public Enterprises and co-financed by the European Union.
The photo shows (from left to right) Jan-Hendrik de Wiljes (postdoc committee member), Ralf Borndörfer (promoter and committee head), Felix Prause, Natalia Kliewer (committee member), and Nicolas Perkowski (committee member).
Link to BerDiBa project:
https://www.zib.de/research/projects/berliner-digitaler-bahnbetrieb
Congratulations!