Abstract
This article presents a robust and efficient algorithm for real-time fault detection (FD) of neutron flux detectors, a critical component in instrumentation and control (I&C) systems of nuclear power plants (NPPs). The proposed methodology addresses the challenge of sensitive FD index generation for nonlinear Gaussian systems by leveraging an enhanced sequential Monte Carlo (SMC) variant. These detection indices are processed using a novel statistical divergence measure, the gradient-based Kullback–Leibler divergence (GKLD). To enhance robustness and sensitivity to incipient faults in the presence of high noise levels, the GKLD output is integrated with an exponentially weighted moving average (EWMA) framework. A comprehensive sensitivity analysis is conducted to evaluate key tuning parameters impact, such as the EWMA smoothing factor ( $\alpha $ ) and threshold multiplier (h), on the probability of missed detection (PMD) and probability of false alarm (PFA). The results, including an ablation study quantifying the contribution of individual components and extended Kalman filter correction-based particle filter (EKF-PF, GKLD, and EWMA) to overall performance, confirm the robustness and effectiveness of the proposed FD method. The findings also highlight the method’s ability to handle varying fault-to-noise ratios (FNRs) and demonstrate its applicability to real-time reactor FD systems. To validate the effectiveness of the proposed framework, pressurized heavy water reactor (PHWR) neutron flux variations driven by reactivity changes are modeled using point kinetic equations (PKEs), which accurately represent reactor dynamics for realistic conditions and measured by neutron flux detectors, distributed across the reactor core to provide critical real-time information about core status.

This publication has 30 references indexed in Scilit: