A physics-informed neural network-based edge-cloud collaborative digital twin for predictive maintenance of ageing railway signalling equipment

This study proposes a physics-informed neural network (PINN)-based edge-cloud collaborative digital twin (DT) for predictive maintenance of ageing railway signalling equipment (RSE).The core model, an adaptive pigeon-inspired optimised PINN (API-PINN), integrates electromechanical degradation physics with operational data to forecast faults and estimate remaining useful life (RUL).The architecture enables low-latency, real-time anomaly detection at the network edge while leveraging the cloud for intensive model training.Experiments on real-world railway signalling data sets can show that API-PINN attains the prediction accuracy of 98.77% and the R 2 of 0.968, which is better than the baseline models.The given framework offers a strong performance when the data is noisy and imbalanced, allowing for the arrangement of the maintenance schedule proactively, minimising the costs of operations, minimising downtimes, and improving safety.