Self-powered and machine learning assisted low-energy interpretable AI system for freight train bearing condition monitoring

Current freight wagon bearing monitoring primarily relies on wayside acoustic and temperature detection systems deployed along specific track sections, supplemented by periodic manual inspections at stations, as onboard power is unavailable. However, such intermittent monitoring fails to deliver timely or accurate assessments of bearing health. To address these limitations, this study proposes a low-power artificial intelligence (AI) monitoring framework that seamlessly integrates an energy-harvesting frontend with a physically interpretable fault-analysis backend, enabling real-time bearing health monitoring. A data-driven dynamic frequency method is introduced to construct phase-plane signals, facilitating rapid and precise identification of bearing fault signatures. The principal advantage of this approach lies in its ability to establish an accurate mapping between fault characteristics and changes in physical parameters associated with bearing conditions, supporting low-power edge-level information processing. Experimental results demonstrate that the proposed system achieves stable self-powered operation and autonomous fault diagnosis under conditions representative of real freight wagon service. Moreover, the integration of energy harvesting with physically interpretable fault-analysis algorithms achieves low-energy interpretable artificial intelligence (Low-Energy + I + AI), advancing beyond traditional energy harvesting and artificial intelligence (Energy + AI) concepts, thereby driving significant advancements in intelligent bearing monitoring technologies for railway transportation.