Ballastless Track Support Deterioration Evaluation Using Machine Learning

Ballastless tracks have been widely used for highspeed rail systems globally since their maintenance is relatively minimal. However, support deterioration right beneath the in-between slabs' connectors has been usually reported and quite well known in the industry. Anywater ingress can quickly undermine the condition of cement-stabilized soil that supports the track slabs. It is thus very crucial to very early detect the impaired condition of the slab supports since mudded support can result in poor ride quality and eventually endanger highspeed train operations. Therefore, the ability to predict the deterioration of track slab supports is highly beneficial to predictive and preventative maintenance in practice. In this study, track slab support stiffness is considered as a precursor to identify the severity of deterioration. The nonlinear FE models, which were validated by field measurements, have been used to populate data in order to develop machine learning models capable of evaluating the track support deterioration. Axle box accelerations are adopted in a form of datasets for machine learning models. Parametric studies have yielded a diverse range of datasets considering the train speed variations, train axle loads, and irregularities. The results demonstrate that the machine learning models can reasonably diagnose the condition of the track slab supports. The outcome reveals the potential of machine learning to evaluate ballastless track support deterioration in practice, which will be beneficial for railway maintenance.