Train marshalling is an essential step in railway operations, where the cars on the train are arranged in a way that is operationally feasible and also mitigates the risk of derailment. Many railway operators have implemented train marshalling rules based on a combination of experience and simulations using ranges of assumed input parameters. There has been a renewed focus on train marshalling in light of recent events, which has gained both industry and regulatory attention. Rules-based train marshalling is easy to implement, but can be inadequate in cases where a more nuanced assessment is necessary, incorporating actual vehicle and track conditions. However, since train marshalling is a real-time process, it does not allow detailed multi-body dynamics simulations to be carried out individually due to the time MBD simulations require with currently available computational power. As the implementation of wayside and mobile measurement technology becomes widely adopted by railway operators and infrastructure owners, there is an opportunity for better train marshalling by synthesizing near-real-time measurement data. This thesis demonstrates the need for enhanced train marshalling using theoretical and real-life scenarios, where diferent decisions could be made if the current conditions of the vehicle and track are understood. The investigations are then carried out using multiscale simulations to assess the infuence of wheel and rail profles, track geometry, train makeup, and train handling on derailment risks. An initial assessment of the efectiveness of laser profle measurements against measurements collected using a contact-based proflometer is also conducted. Regression and classifcation techniques are explored to develop potential metamodels that heavily rely on the insights from measurement data and simulation results while allowing the train marshalling process to be carried out promptly. A framework for an enhanced train marshalling process is proposed, combining data synthesis, multiscale simulations, and metamodeling into a structured process that can be further expanded as research into individual areas of the framework continues to advance.