Generating realistic track irregularities using a diffusion model adapted for time series data

Vehicle System Dynamics, vol. 64, pp. 2104–2117

Abstract

A dataset of appropriate track irregularities is essential for the dynamic simulation of rail vehicles. Their characteristics can significantly impact wheel/rail forces, ride comfort, and load assumptions. When measured track irregularities are unavailable, or when consistent track quality is required throughout the simulation, synthetic track irregularities are often used. These irregularities are typically generated using an inverse Fourier transform. However, track irregularities generated by this method do not exhibit the same characteristics as real measured data and the usage can have several drawbacks. To address these issues, a novel approach using artificial intelligence is demonstrated in this work. A denoising diffusion probabilistic model (DDPM) is adapted for time series and then trained on a large dataset of measured track irregularities. Based on the learned representation of the data, new and unique track irregularities can be synthesised from pure noise. The track irregularities produced by this method exhibit more realistic features compared to those generated by the traditional inverse Fourier method.

Authors 4

1. Philipp Leibner corresponding Aachen Institute for Rail Vehicles

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University

   Institute for Rail Vehicles, RWTH Aachen University, Aachen, Germany

2. Aaron Ketteler Aachen Institute for Rail Vehicles

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University

   Institute for Rail Vehicles, RWTH Aachen University, Aachen, Germany

3. Christian Schindler Aachen Institute for Rail Vehicles

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University

   Institute for Rail Vehicles, RWTH Aachen University, Aachen, Germany

4. Raphael Pfaff Aachen Institute for Rail Vehicles

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University

   Institute for Rail Vehicles, RWTH Aachen University, Aachen, Germany