AI–Digital Twin framework for automated predictive maintenance of bonded insulated rail joints

Predictive maintenance is a key enabler of automated Structural Health Monitoring (SHM) for safety-critical infrastructure, supporting reliable operation under increasing service demands. This paper presents an integrated Digital Twin–Artificial Intelligence (DT–AI) framework for the automated condition assessment of bonded Insulated Rail Joints (IRJs) and adjacent sleepers. A high-fidelity Digital Twin is developed to simulate joint behaviour under varying preload and support degradation scenarios, generating physically consistent datasets across representative loading conditions. Deep Convolutional Neural Networks (DCNNs) and traditional Machine Learning (ML) classifiers are trained on time–frequency representations of selected predictive indicators to identify and classify degradation states. Results demonstrate that the proposed physics-informed DT-driven framework achieves high diagnostic accuracy and strong generalization capability, with DCNNs models outperforming conventional classifiers. By integrating physics-based simulation with automated data analytics, the proposed framework advances scalable, data-driven SHM and predictive maintenance strategies, for the intelligent automation of railway infrastructure management.