Prognostics and health management via long short-term digital twins

Current digital twin-based Prognostics and Health Management (PHM) research mainly focuses on prediction with a few parameters or a single event. However, when the relationship between moving parts of equipment is complex, both instantaneous failure and long-period degradation should be considered. Existing research is challenging to describe the dynamic evolution of the health status of the target object at varied time scales. In addition, data characteristics at different time scales are difficult to be captured simultaneously by current methods. This paper proposes an innovative dual time scale digital twin modeling and analysis method. According to the PHM business rules, the time series signals are decomposed into fine-grained scales and adaptively constructed into short time scale and long time scale digital twins. The generated events of different scales pay attention to the temporal characteristics and uncertainties, and interactive mapping of events at different scales is realized in cyberspace. Events at a short time scale focus on the real-time occurrence of anomalies, and longterm events track equipment degradation and trends. The interaction and collaboration between different time scale models are also discussed. Finally, the paper uses the state monitoring of large cranes in iron and steel enterprises to verify the proposed method. The results show that this modeling method can reduce the uncertainty and incompleteness of system monitoring in a complex system. Real-time performance and reliability of equipment health diagnosis have been effectively improved.