Keywords: smart monitoring, artificial intelligence, probabilistic modelling, predictive maintenance, railway infrastructure, railway track, dynamic multi-regression hidden Markov model, remaining useful life.

Smart monitoring technologies and artificial intelligence are increasingly deployed in railway systems, generating large volumes of operational data from in-service infrastructure. However, translating these heterogeneous and indirect measurements into actionable maintenance decisions remains challenging, particularly because railway track degradation processes are partially observable, non-stationary, and influenced by complex interactions between traffic loading, environment, and maintenance actions. This paper presents a degradation-aware modelling framework for railway infrastructure that integrates wayside condition monitoring data with probabilistic state-space degradation modelling. The framework is formulated as a decision-relevant analytical engine that can be coupled to existing railway digital twin environments, rather than as a standalone or geometric digital twin. Asset condition is represented through latent degradation states inferred from axle-level dynamic response measurements, operational variables, and environmental factors using a Dynamic Multi-Regression Hidden Markov Model. The inferred degradation states are linked to state-conditioned hazard models and Remaining Useful Life estimation, enabling monitoring data to be translated into uncertainty-aware risk metrics suitable for maintenance planning. A case study focusing on rolling contact fatigue demonstrates the ability of the proposed framework to identify degradation regimes and decision-relevant risk windows consistent with operational practice. The results highlight the value of probabilistic degradation modelling as an enabling component within digital twin-based railway asset management systems.

download the full-text of this paper (PDF, 12 pages, 870 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description