Modelling component failure rates utilizing sensor-based degradation data

In the nuclear power industry, risk monitors are intended to provide a point-in-time estimate of the system risk given the current plant configuration. Current risk monitors are limited in that they do not take into account the deteriorating states of plant equipment. Current approaches to on-line risk monitors use probabilistic risk assessment (PRA) techniques, but the assessment is typically a snapshot in time. Living PRA models attempt to address limitations of traditional PRA models in a limited sense by including temporary changes in plant and system configurations. However, information on plant component health (a.k.a. level of degradation) are not considered. This often leaves risk monitors using living PRA models incapable of conducting evaluations with dynamic degradation scenarios evolving over time. There is a need to develop enabling approaches to enhance risk monitors to provide time- and condition-dependent risk by integrating traditional PRA models with condition monitoring and prognostic techniques. This paper presents an exponential model for estimating the mean failure rate of components undergoing degradation, and Bayesian inference for updating the distribution of component failure rates. Such a degradation model is based on component performance data gathered over the service life and historical failures. The proposed model is demonstrated using component performance and failure data obtained for five motors subjected to accelerated degradation. The model provides a more realistic picture of the component risk and also forms an important prognostic tool capable of aiding risk-informed decision making.