Track buckling predictions using long track dynamics models and machine learning

Railway track buckling is a critical structural failure phenomenon where continuous welded rails (CWR) deform laterally due to thermal stresses, dynamic loads, or a combination of both. This instability can compromise the track structure, leading to severe safety risks and operational disruptions. Understanding the mechanisms behind track buckling is essential for developing effective strategies to predict, prevent, and mitigate such failures. Despite extensive studies in existing literature, opportunities remain for further investigation into comprehensive simulations of long track dynamics. While virtually infinite track dynamics simulations exist and provide valuable insights, they do not fully represent extended continuous track infrastructure systems, as they are based on assumptions that may not capture the full range of complexities present in such systems. Furthermore, the interplay between dynamic forces and buckling behaviour over realistic track lengths remains an area where simulation studies are relatively scarce, indicating opportunities for further investigation. This thesis advances the field by developing computational models that integrate advanced simulation techniques with predictive analysis. Its aim is to establish a cohesive framework for modelling CWR buckling risks by deepening the understanding of track dynamics under combined static, thermal, and dynamic forces. One key objective of this thesis is to develop a physics-based, numerical dynamic railway track model capable of simulating both buckling phenomena and long tracks. Here, “long” refers to track lengths sufficient to accommodate realistic train configurations, typically in the range of 2 to 3 kilometres. The model is designed to leverage the finite element method for rail representation, mass block models for the sleepers and the underlying supports, and parallel computing techniques using the message passing interface framework to efficiently simulate kilometre-scale track dynamics under varied operational scenarios. A further objective is to ensure that the model captures the nonlinear responses of the track system to combined thermal stresses and dynamic lateral loads, enabling representation of track dynamics under diverse loading and environmental conditions. Another central objective is to incorporate uncertainties in critical parameters such as rail temperature, lateral resistance, misalignment, and ballast properties into a probabilistic framework based on Monte Carlo simulations to assess track stability. To manage the computational demands of kilometre-scale simulations, a further objective is to develop a machine learning-based surrogate model that accelerates evaluation while maintaining predictive fidelity. This stochastic approach to buckling risk evaluation enhances the reliability of buckling risk assessments by accounting for variability in dominant buckling drivers and paves the way for real-time predictive capabilities in railway infrastructure management. The thesis synthesises numerical methods with contemporary data-driven techniques, creating a hybrid modelling framework that bridges the gap between physics-based simulations and machine learning. This integrated approach enhances both the predictive capability and operational applicability of the models. This thesis demonstrates the feasibility of kilometre-scale dynamic buckling simulations through parallelised physics-based modelling and establishes a probabilistic framework that captures variability in key buckling parameters. It further shows that the machine learning-based surrogate model achieves strong predictive consistency with the physics-based simulations, exceeding 90% agreement. Together, these advances contribute a unified, scalable, and configurable framework that enhances both the predictive capability and operational applicability of buckling risk assessments. The framework developed in this thesis provides insights for asset maintenance and risk management, enabling safer and more resilient railway operations. By integrating advanced simulation techniques, probabilistic analysis, and machine learning, it delivers new strategies for mitigating track buckling and advances current practice in railway infrastructure management.