MODEL-BASED AND DATA-DRIVEN DIGITAL TWINS FOR RAILWAY VEHICLE-TRACK INTERACTION MONITORING

Abstract

The concept of digital twins is a promising alternative approach for condition monitoring in the railway sector. In this work a digital twin for the vehicle-track interaction is presented, which is based on a physics-informed encoder-decoder convolutional neural network. The digital twin can be used bi-directionally. On the one hand, it can estimate the longitudinal rail profile from axle-box accelerations and, on the other hand, axle-box accelerations can be estimated from the rail profile. In this way, the model can be used to get actual information on the condition of the rail from onboard data and additionally to predict vehicle reactions to a specific rail longitudinal profile. The model is trained and tested on real data acquired with a shunter locomotive.