Advancement of data-driven SHM: A research paradigm on AE-based switch rail condition monitoring

The past ten years have witnessed the tremendous progress of structural health monitoring applications in civil infrastructures. This is particularly embodied in railway engineering. The increasing train speed brings greater challenges to safety and ride comfort, and the primary theme of maintenance has been gradually altered from offline inspection to online monitoring. Rail operators must get an in-time warning of potential structural defects before critical failure takes place. It is more favourable that the rail operators can take hold of the real-time status of the key components and infrastructures in railway systems. This paper summarizes a long-term research series by the authors’ research team on online monitoring of rail tracks at turnout areas utilizing acoustic emission-based sensing technique, and more importantly, successively advancing signal processing methods and data-driven analysing frameworks, covering Bayesian inference, convolutional neural networks, transfer learning and task similarity analysis. The proposed algorithms tackle noise interference brought by wheel-rail impacts, great uncertainties in an open environment, and insufficiency of monitoring data, and realize comprehensive monitoring of rail tracks in turnout areas from basic crack detection to regressive condition assessment step-by-step.