AI powered digital twin for rail damage prediction

Rail digital twins (DT) can enhance efficiency in railway operations as they have the potential to generate more accurate predictions on rail degradation. Increased accuracy is achieved by DTs through combining longitudinal train dynamics, multibody dynamics, traction mechatronics, material damage, and friction, among other models. These models are computationally intensive and are not feasible for routine railway operations, which involve constant variations in train configuration, vehicle types, speed restrictions, etc. This paper introduces an AI leveraged railway DT to predict rail damage along a railway track. The DT is implemented through longitudinal train simulations (LTS) and multiphysics vehicle multibody simulations (MBS) to determine rail damage. These rail damage results are then used to train a machine learning algorithm surrogate model to predict rail damage using inputs available onboard the locomotive. The proposed framework generates rail damage predictions in real time, without the need of rerunning complex simulations for different operational scenarios.