Smart fault monitoring in long railway tracks using acoustic emission signals and dynamic neural network

Railway infrastructure must be maintained properly to ensure safe and efficient rail service. Conventional track monitoring methods, relying on manual inspections and scheduled maintenance, often fail to detect faults in a timely manner, increasing the risk of operational disruptions and accidents. Acoustic Emission (AE) sensing offers an effective approach for rail monitoring by capturing high-frequency stress waves associated with crack initiation and propagation. However, conventional Artificial Neural Network (ANN) models used for analysing AE data exhibit limited performance over longer rail spans due to environmental variability and complex field conditions. This study investigates the suitability of Dynamic Neural Network (DyNN) modelling for fault detection in extended rail sections. A novel DyNN architecture is developed to capture the temporal characteristics of AE signals and is validated using real world field data collected from a 12 m rail span. The study reviews existing AI-based rail fault detection methods and evaluates the DyNN model through laboratory and field experiments. Results demonstrate that while ANN models perform well for short spans, the proposed DyNN model achieves superior fault localisation accuracy for longer spans, with prediction errors below 1%. These findings highlight the potential of DyNN-based models for scalable railway health monitoring.