Solid mechanics and artificial intelligence hybrid approach to mitigation of climate change driven railway track buckling

Mitigating railway track buckling risk is a crucial aspect of keeping rail travel safe during summer heat and relies on prediction of sites at risk of buckling to direct mitigation efforts. Models capable of predicting buckling stability are available; however, their application to real-world sites is made challenging due to uncertain track parameter information. This thesis investigates the development of a hybrid methodology capable of addressing this challenge and combines analytical models, experimental data and a fuzzy inference model to predict the buckling stability of track using linguistic data. A fuzzy inference model, extending the capabilities of a previous model, was developed and shown to be effective for predicting buckling stability for input track parameters. The model was further developed through application to an unrelated set of recorded data, showing the effectiveness of the methodology for non-unique training datasets and real-world application. An experimental investigation established lateral resistance values for sleeper displacements in ballast critical to buckling prediction. A previously unreported phenomenon of sleeper uplift during single sleeper push tests was examined, finding a strong link between uplift restraint and lateral resistance, particularly for steel sleepers. Linguistic descriptions of ballast conditions were used to model track lateral resistance uncertainty, establishing fuzzy sets through a combination of results from track expert interviews and experimental data. The methodology was seen to be successful in predicting buckling stability when applied to data describing historical buckles and predicted higher risk than a comparison alternative method using averaged lateral resistance values. The application of the hybrid methodology has proven to be an effective technique of predicting buckling risk using uncertain track parameter information, enabling comparison of realistic mitigation measures. Furthermore, the methodology is seen to be versatile and transferrable, with application to a wide range of potential applications beyond the field of track buckling.