Implementation of an AI-based predictive structural health monitoring strategy for bonded insulated rail joints using digital twins under varied bolt conditions

Abstract Predictive maintenance is essential for the implementation of an innovative and efficient structural health monitoring strategy. Models capable of accurately interpreting new data automatically collected by suitably placed sensors to assess the state of the infrastructure represent a fundamental step, particularly for the railway sector, whose safe and continuous operation plays a strategic role in the well-being and development of nations. In this scenario, the benefits of a digital twin of a bonded insulated rail joint (IRJ) with the predictive capabilities of advanced classification algorithms based on artificial intelligence have been explored. The digital model provides an accurate mechanical response of the infrastructure as a pair of wheels passes over the joint. As bolt preload conditions vary, four structural health classes were identified for the joint. Two parameters, i.e. gap value and vertical displacement, which are strongly correlated with bolt preload, are used in different combinations to train and test five predictive classifiers. Their classification effectiveness was assessed using several performance indicators. Finally, we compared the IRJ condition predictions of two trained classifiers with the available data, confirming their high accuracy. The approach presented provides an interesting solution for future predictive tools in SHM especially in the case of complex systems such as railways where the vehicle–infrastructure interaction is complex and always time varying.