Assessment of algorithms for prognostics of fatigue damage initiation in operating nuclear power plants

Aging management of components in nuclear power plants (both legacy plants as well as advanced light water reactors) is important to ensure safety and economic operation over the lifetime of the plant. Active and passive components in nuclear power plants, in the presence of elevated temperatures, pressures and irradiation, experience degradation that can impact their ability to perform their functions. Managing degradation in these components requires the ability for early detection of degradation-ideally before the level of degradation reaches a state where component replacement is the only viable option. To improve the options available for mitigation and management of degradation, a focus on detecting precursors to degradation is desirable. Such precursors might include both early indicators of component condition (to detect incipient degradation) as well as measurements of environmental conditions to better understand the stressors contributing to degradation. These measurements will need to be deployed in a manner that enables continuous condition monitoring to enable rapid detection of degradation. In parallel to the measurement process, methods for analyzing the measured data to identify component condition (i.e., diagnostics) and algorithms for estimating remaining useful life (RUL) of components and systems based on the degradation information (prognostics) are needed. The assessment of remaining life is important in proactive- or prognostic-based life management of these facilities because such "condition-based maintenance" strategies can potentially improve safety and reduce costs by detecting damage and scheduling appropriate mitigation strategies early in the component lifecycle. This paper presents an evaluation of prognostic algorithms for RUL estimation in degraded passive components in nuclear power plants. Specifically, three Bayesian prognostic approaches, namely the extended Kalman filtering, unscented Kalman filtering, and particle filtering, are evaluated for their ability to predict initiation of fatigue cracking in stainless steel structural components. Empirical models of degradation accumulation based on non-linear ultrasonic measurements are used to capture damage evolution in austenitic stainless steel before initiation of fatigue cracking. Experimental data are used in the evaluation, and metrics used for the comparison include prediction accuracy and computational cost.