Proposal of an online univariate monitoring algorithm for sensors in primary circuit of nuclear reactor based on sliding-window kernel density estimation and quantitative metrics

The reliability and accuracy of safety-related sensors in the primary circuit of a nuclear reactor are crucial to ensure the safety and economic viability of power plants. In recent years, substantial research has focused on the development of online monitoring (OLM) and condition-based maintenance (CBM) methods to enhance conventional maintenance strategies that rely on periodic manual calibration (PMC). However, certain measurements from the power plant instrumentation system exhibit weak correlations with other variables. These uncorrelated signals are categorized as either independent or isolated signals. Although these signals are unsuitable for inclusion in multivariate-based OLM systems, they still require monitoring. This study introduces a novel data-driven algorithm based on sliding-window kernel density estimation (SWKDE) combined with quantitative metrics for online statistical univariate monitoring. The effectiveness of the proposed method was demonstrated by detecting sensor faults using data retrieved from the CPR1000 full-scope pressurized water reactor (PWR) simulator. Results of the case studies indicated that typical types of sensor soft faults, including bias, drift, and noise faults, can be successfully detected by identifying data anomaly patterns using qualitative metrics. • An online univariate monitoring algorithm based on sliding-window kernel density estimation (SWKDE) and quantitative metrics was proposed. • This approach aims to fill an important gap in the uncorrelated univariate online monitoring of sensors in nuclear power plants.