Big data analysis and multi-sensor fusion for fault diagnosis in high-speed railways

Staff Mentor: B. De Schutter (Bart)

Other Mentor(s):

Dr. A.A. Nunez, Dr. S. Faghih Roohi


Railway networks; Machine learning; Identification and estimation


The Netherlands is connected to the European high-speed rail network with railway lines that are being monitored regularly for fault detection, maintenance works, and repairs. There are several types of monitoring systems and measurements such as ultrasonic measurements, video image systems, geometrical measurements, eddy current measurements, and ARROW for rapid measurement of sound, roughness and defects of railway tracks. In Figure 1, some images of rail defects are presented. For the classification of these defects in several severity levels and prediction of the growth of faults, such as small cracks becoming deeper, learning algorithms such as multi-layer neural networks can be used. We are looking for an efficient algorithm to diagnose, classify, and predict faults.

The aims of this M.Sc. project are:

• Identification and classification of faults in railway tracks by data analysis

For assessing track conditions of high-speed lines, there is a huge amount of data from different measurement systems/sensors. The approach is to show the usability of the collected data in sensor fusion and maintenance decision making of railway tracks. This can be done by the identification, classification, and prediction of future or near-to-happen defects from the trained and classified datasets. Here, the aim is to develop methods to classify and label the current fault data, which may be originating from different measurement sensors.

• An efficient learning approaches for sensor fusion

Integration of data collected from different measurement sensors with various characteristics is a challenge that should also be addressed in this project. To solve this challenge, deep learning approaches and artificial neural networks are powerful tools for modelling, predicting and classifying high-dimension multi-source data.

This research is in collaboration with Infraspeed, the builder and infrastructure maintenance company of the HSL-Zuid high-speed line in the Netherlands.

Representation of a multi-layer neural network for fault diagnosis and classification in a highspeed rail track

© Copyright Delft Center for Systems and Control, Delft University of Technology, 2017.