A Machine-Learning-based Surrogate Modeling Methodology for Submodel Integration in the Holistic Railway Digital Twin Platform

Abstract

A holistic railway infrastructure digital twin (DT) platform is sophisticated and consists of a series of submodels (e.g., turnouts, tracks, vehicles, etc.) that are built through various methodologies and software. However, integrating these submodels into the DT platform is tremendously challenging due to considerable computational complexity, software and interface restrictions. To this end, we designed a machine learning (ML) based surrogate modeling methodology for the submodel integration in the holistic railway infrastructure DT platform and illustrated the methodology through a case study. In this case study, an ML-based surrogate model for multibody simulation of railway vehicle-track dynamics is created, which can replace the railway vehicle-track simulation executed with the Multibody Dynamics (MBD) Simulation commercial software SimPACK. The well-built ML model can accurately and quickly predict the vehicle-track system's dynamic responses to different track irregularities. Besides, the integration process of the ML-based surrogate model into the DT platform through a standardized open-source Functional Mock-up Interface (FMI) is also proposed. The developed surrogate modeling methodology shows great promise owing to its high fidelity, which is verified by the measurement data collected from the Austrian national railway track system. The main contribution of our work lies in the well-built ML-based surrogate modeling methodology for reducing the computation complexity and time of different submodels, which facilitates the unification and integration of different submodels. Furthermore, this approach can also be applied to other submodels and help to build the holistic railway DT platform collaboratively.