Latent fault detection and diagnosis (LFDD) for equipment are crucial for safety and reliability in nuclear power reactors( NPRs). In this paper, a high accuracy LFDD method combining gate recurrent unit based autoencoder (GRU-AE) and Random Forest (RF) was proposed for control rod drive mechanisms (CRDM) in pressurized water reactors (PWRs). The movement sequences of the CRDM coils’ current reflecting the degradation of CRDM were assumed as time series and taken by the sliding window. GRU-AE, a network containing an encoder and decoder, was applied to build the health operating data reconstruction model. The labeled abnormal data was generated by adding the typical current waveforms with a gaussian noise sequence. The latent fault could be detected by feeding online datasets into the well-trained model because of reconstruction error sensitivity to a slight/minor deflection. And then, RF was used as a classifier to diagnose the types of abnormalities. The results demonstrate that combining the GRU-AE and RF can realize high-accuracy detection on real-world imbalanced datasets.

The condition assessment of equipment in nuclear power plants (NPPs) could provide essential information for operation and maintenance decisions, which would have a significant impact on improving the safety and economy of NPPs. To date, substantial work has been conducted on the condition assessment based on machine learning for NPP equipment. To provide a comprehensive overview for researchers interested in developing machine learning methods for NPP equipment condition assessment, this critical review presents a detailed literature survey on state-of-the-art research and identifies challenges for future study. Valuable information is presented, including major failure modes, data sources, maintenance strategies, and the relationship between equipment lifetime, assessment technology, and maintenance strategy. Following the typical lifetime of NPP equipment for condition assessment, current works in this domain are categorized into anomaly detection, remaining useful life prediction, and fault detection and diagnosis. The techniques and methodologies adopted in the literature are summarized from each aspect. In particular, the in-depth NPP equipment condition assessment survey based on deep learning methods is presented. In addition, we elaborate on current issues, challenges, and future research directions for the condition assessment of equipment in NPPs. These directions we believe will pave the way for equipment condition assessment.

Neutron detectors in nuclear power plants (NPPs) are critical for system stability, yet their malfunctions may lead to false alerts and misdiagnoses. Multidetectors deployed in diverse positions vary with the nuclear reactor states contained spatial-temporal variations of neutron fluxes. Existing methods seldom concurrently consider intricate spatial-temporal correlations and gradual state variations among detectors. This study proposes a detector-oriented fault detection and isolation method named the spatial-temporal state adaptation model (ST-SAM). The method introduces a local-global spatial-temporal network that captures the potential interdependencies within the detector topology. To minimize cross-state discrepancies in reactors, ST-SAM integrates three submodules: a signal reconstructor to enhance the specific-state variation representation; a correlation alignment to mitigate interstate feature discrepancies; and an adversarial discriminator to extract spatial-temporal state-invariant features. Leveraging the parallel detection strategy, ST-SAM effectively detects and isolates faulty detectors, preventing fault propagation on subsequent diagnosis. Experiments on ex-core and in-core neutron detectors in real-world NPPs with simulated faults verify that the ST-SAM outperforms various state-of-the-art methods in terms of signal reconstruction and fault detection.

Balanced in-core power levels in nuclear power plants (NPPs) are critical for safety, whereas power tilt disrupts this balance, reducing safety margins and posing risks.0Early warning for power tilt offers an effective way of optimizing monitoring. Due to abnormal-sample scarcity and security concerns, common data-driven models train on the simulated data generated by simulators. However, achieving a satisfactory effect in practices is difficult because simulators imperfectly emulate reality. Thus, we propose a power tilt-oriented early warning method called simulation-reality spatial-temporal model (SR-STM).0Motivated by the physical model in NPPs, a knowledge-guided hierarchical graph is designed to characterize spatial correlations among local power levels for SR-STM’s input. The SR-STM uses a lightweight spatial-temporal network (LST-Net) as a feature extractor, balancing precision and efficiency. To bridge sim-real inter-domain discrepancies, SR-STM utilizes node-alignment adversarial learning for fine weight tuning in subdomain, and eigenvalue-based scale alignment for sim-real feature proximity. Forecasting local power levels using the SR-STM, dynamic metrics and alarm limits are calculated and compared to perform the early warning task. The online experimental prototype verifies that SR-STM surpasses various state-of-the-art methods in terms of early warning and sim-real cross-domain tasks.

In nuclear power plants (NPPs), ex-core neutron detectors are deployed around reactor cores and are essential for reactor stability, but their deterioration and malfunction can cause misperceptions and misdiagnoses. Existing fault detection seldom accounts for global spatial-temporal coupling relationships implied among overall detectors and uncertainty under transient operations. Thus, we propose a novel detector-oriented fault detection scheme called the global-fused dynamic detection (GFDD) model, established by the global spatial-temporal graph (GSTG), moving-global graph convolution (MGGC), and uncertainty-quantified dynamic detection (UQDD). To enrich informational sources and disperse faulty propagation, we specifically design the GSTG for characterizing the spatial-temporal relationships among overall detectors and the MGGC for efficiently capturing global high-level features, further generating multi-detector reconstructed signals and residuals. Through calculating dynamic statistics and quantifying uncertainty under varying operating conditions, the UQDD identifies faulty detectors and corrects erroneous signals. Experiments on steady and transient states from a real-world NPP with simulated faults validate that the GFDD model outperforms various state-of-the-art methods with regard to signal reconstruction and fault detection.