Project Manager: Prof. dr. Paul Havinga
applied maintenance strategies in the railway sector are mainly based on past
experience and periodic inspections. However, the inspections do not
effectively prevent unexpected failures of the network, especially since degradation
rates are rather unpredictable. As a result, many maintenance activities are
performed in a reactive way, which yields network unavailability and relatively
high maintenance costs. In a large research project, we now aim to transition
from reactive maintenance to (pro-active) predictive maintenance. This will be
achieved by developing and utilizing a range of continuous monitoring systems
(data capturing) and the associated prediction models (data analytics) for
failures of infrastructure elements (e.g. switches, joints, level crossings,
track section) and by continuously improving failure mode, effects and
criticality analyses (FMECAs) and root cause analyses (RCAs). This will
significantly reduce the uncertainty on the actual condition of the network
elements. The rail network already contains some sensors and there is a
significant amount of data on usage, failures, maintenance activities and
performance available. However, this data does not always contain the most
relevant parameters and is not always processed in the right way, which means
that the amount of useful information obtained from the data is rather limited.
of the project is to develop data processing algorithms and
associated diagnostic and prognostic methods for the different infrastructure
elements, based on all data sources available in the field. This will be
achieved by taking the following steps. Firstly, an overview of all available
types of data must be created. Data will be available from traditional
maintenance-related sources (failures, maintenance activities), but also
additional sources could be identified: passengers and their mobile phones,
satellites, sensors on measuring trains, regular trains and monitoring systems
on the tracks. The second step will be to develop methods for collecting and
merging all different data types. The third step, which constitutes the core of
this project, is to develop data processing methods that retrieve information on the actual state
(diagnosis) or expected future failures (prognosis) of the railway elements.
The ‘big data’ must be analyzed, where methodologies like machine learning,
neural networks, data mining techniques, pattern recognition, etc. should be
investigated. In a parallel PhD project within the same program, a physical
model based approach will be followed, using the same railway elements and data
sources. This will provide a unique opportunity to compare the two approaches.
Moreover, the gained insight in the system and failure behavior also may guide
the data mining, which makes it more than only a blind identification of the failures.
On successful completion of the third step, the final step will be the
application of the retrieved information in optimizing the maintenance process.
If failure probabilities and life time distributions of critical elements can
be quantified, these can be used as input for maintenance models, yielding
decision support for the planning and clustering of maintenance tasks.
This project is part of a long-term collaboration
between Strukton Rail Maintenance and the University of Twente.
Project duration: 1 April 2015 - 1 April 2020
Participants: STRUKTON, UT
Project budget CTIT: 530 k-€ / 220 k-€ funding
Number of person/months CTIT: 1.2 fte/year