Track geometry monitoring using measured data from commercial trains towards predictive maintenance

Recognizing track problems is a crucial responsibility for railway engineers to ensure safe train operations. Analysing track geometry (TG) characteristics is essential for developing an effective track condition monitoring strategy. Recent studies have focused on identifying track geometry irregularities using onboard vehicle dynamics data, with a particular emphasis on vertical irregularity. This is because it is easier to understand the relationship between vehicle reaction and acceleration in the vertical direction, and there is a straightforward method for identifying vertical irregularities. In contrast, lateral irregularities are more challenging to detect due to factors such as wheel conicity, non-linearity of wheel-rail contact, and the Klingel hunting motion. Addressing the problem of lateral irregularities requires tracking the contact point between the wheel and the rail in the lateral direction. This research focuses on detecting lateral irregularities by identifying the relationship between lateral acceleration and the lateral displacement of the wheel in relation to the rail (LDWR). The first part of this thesis is related to the activities of the EU project Assets4Rail, funded within Shift2Rail in which a method for detecting LDWR has been established by developing a sensor system as an alternative to inertial platforms, and developing the image processing algorithms for analysing the video output data. In addition, the next part of this research uses Machine Learning (ML) models to solve the non-linearity issue for detecting the lateral irregularities by LDWR. The study employs a supervised ML model trained and tested with numerical simulation results from Simpack, and Gensys software for straight and curved sections of various tracks. Testing different algorithms allows for identifying the best models for this purpose and determining the most efficient results. Hence, this work summarizes the methodology and results of establishing a method for finding the relationship between the lateral acceleration and lateral irregularities from the lateral displacement of the wheel relative to the rail measured on board and proposing an algorithm to detect the lateral irregularities of the track. The conclusion of this thesis can lead researchers to predictive maintenance in further studies by developing algorithms for an on-board sensor system monitoring capable of detecting LDWR and detecting lateral irregularities Overall, this research highlights the significance of addressing lateral irregularities in track maintenance to ensure safe and efficient train operations.