A framework for state assessment and remaining useful life prediction of control rod drive mechanism roller

Abstract

The control rod drive mechanism (CRDM) roller is susceptible to degradation, such as wear and fatigue, under environmental and operational stresses, including temperature, humidity, friction, and impact. To enable timely operating state diagnosis, predict future degradation trends, and support operation control and maintenance decisions, this study develops a framework for CRDM roller state assessment and remaining useful life (RUL) prediction. In the state assessment stage, residual distribution analysis combined with adaptive neighbourhood radius clustering is employed to evaluate roller states under limited input data. In the RUL prediction stage, a particle filter-based degradation model is constructed, with parameters estimated via maximum likelihood and multi-objective optimization, and further corrected using Bayesian theory and backward smoothing to enhance prediction accuracy. Validation on bearing datasets achieved a degradation onset estimation deviation of less than three time steps and a RUL prediction error of 8.19%. At the same time, the application to the CRDM roller yielded an error of 8.96%. These results confirm the framework’s capability for accurate state diagnosis and precise RUL prediction using real-time monitoring data.