Building a digital twin for a train requires a real-time simulation model that accurately replicates the train's behaviour and performance under various operating conditions, enabling digital twin-based applications such as real-time control and monitoring, predictive maintenance, and the prediction of potential faults while testing coping strategies. To meet this requirement, we propose a physics engine-driven approach, employing a physics engine as a real-time simulation tool to model and simulate the dynamic behaviours of the train. We have modelled and simulated the vehicle derailment scenarios using the physics engine-driven method and the traditional multi-body (MB) model, respectively. Comparative analyses between the two types of models have been conducted to assess and validate the feasibility and usefulness of the proposed method. In addition, two application cases of the physics engine-driven method are also presented in the paper. The results demonstrate that while the physics engine-driven model exhibits relatively lower accuracy compared to the traditional MB model, it presents excellent real-time performance. This suggests that if future improvements in accuracy are achieved, such as integrating with the AI model, the newly proposed method could serve as a promising alternative for developing a digital twin train.