Long-horizon multivariate forecasting of nuclear power plant parameters becomes increasingly challenging as the forecast horizon and the number of target variables grow. This study evaluates an long short-term memory (LSTM) autoencoder-based latent trajectory forecasting framework for operator-support-oriented prediction using compact nuclear simulator data. Instead of directly regressing the full future trajectory in the observation space, the proposed framework compresses future trajectories into low-dimensional latent coordinates and predicts those coordinates from short observation histories. Three repeated-seed experiments were conducted to compare direct and latent forecasting across forecast horizons and latent dimensions. Within the direct LSTM baseline considered in this study, direct observation-space forecasting remained competitive at short horizons; whereas, latent trajectory forecasting became increasingly advantageous as the horizon increased. In the 1500-step, 25-channel task, the latent model reduced root mean square error (RMSE) by approximately 16% compared with the direct LSTM baseline while using substantially fewer parameters. Latent-dimension sweeps showed that the smallest bottleneck can have limited representational margin, whereas moderate-to-large dimensions form a broad practical operating range rather than a single universal optimum. These findings suggest that long-horizon nuclear plant parameter prediction can be treated as a conditional trajectory-generation problem in which a short recent plant history is mapped to a compact future-trajectory coordinate, providing a practical, parameter-efficient design direction for AI-assisted plant monitoring and operator support.

This review intends to provide a systematic overview of the recent advancements in the risk assessment of safety-critical digital instrumentation and control systems. Given their increasing importance of the risk-informed regulation and applications, the realistic and precise evaluation of the risk associated with the digitalized safety function plays a crucial role in both existing fleet digitalization and new nuclear plant designs. Considering unique characteristics of microprocessor-based digitalization, the precedent research results and the issues related to modeling techniques are discussed. Fault-tolerance mechanism effectiveness and the software reliability of microprocessor-based systems are selected as major research issues in the digital I&C probabilistic safety assessment, thus an overview of practical quantification methods for these factors is provided. Current states of several issues related to the integration into the existing plant PSA models, such as modeling framework, dynamic modeling, software-induced CCF modeling, human reliability modeling, and system-theory-based approaches are also addressed.

As instrumentation and control (I&C) systems in nuclear power plants (NPPs) have been digitalized, concerns about cyberattacks have risen. As part of multiple barriers for securing plants from cyberattacks, process data monitoring has been researched to prevent degradation of situational awareness among main control room (MCR) operators, particularly due to false data injection attacks. With recent advances in artificial intelligence, autoencoders can be utilized for this purpose; however, approaches relying solely on reconstruction error may overlook manipulations when the reconstructed trend follows the manipulated trend. To improve detection sensitivity, especially when false positive alarms must be minimized, this study proposes a latent distance-augmented scoring method that leverages both reconstruction error and latent space distance. The proposed method stores latent vectors of training data in a memory bank and computes k-nearest neighbor distances for query inputs. These two components are combined using Z-score normalization. Experiments using emergency operation data from an NPP simulator and 10 synthetic false data injection scenarios demonstrate that the proposed combined scoring method outperforms both reconstruction error and latent distance approaches at high specificity levels required for practical applications, achieving 26.3% higher sensitivity (0.7644) than reconstruction error alone (0.6053) at a specificity of 0.9999.

Physics-based nuclear power plant (NPP) simulations including thermal-hydraulic system codes require significant computational resources, limiting their applications for real-time prediction during accidents or more exhaustive safety assessments. While data-driven surrogates have been developed to address this challenge, two critical limitations persist: restricted time horizons (typically less than several hundred steps) and lack of predictive uncertainty quantification. This study proposes a novel two-stage deep learning model inspired by latent imagination techniques from reinforcement learning and incorporating Monte Carlo dropout for uncertainty estimation. The model was validated using simulation records of critical safety parameters—peak cladding temperature and reactor pressure vessel water levels—during NPP accident scenarios. Training data were deliberately limited in number to assess the model's ability to estimate prediction intervals representing uncertainty. Results on test data demonstrate that the model can predict parameter trends for both variables with average mean absolute errors (MAE) of 22K and 0.53m over more than 1,700 time steps while appropriately quantifying uncertainty, achieving prediction interval coverage probabilities (PICP) of 98.76% and 93.35%, respectively. In addition, we validated that the quantified uncertainty properly correlated with training data abundance and trend variability. This research contributes to developing advanced surrogate models for safer and more economic operation of NPPs, including small modular reactors.

Long-horizon prediction of plant parameter trends has recently attracted increasing attention in nuclear engineering because it can provide operators with look-ahead information under abnormal transients. Building on this research trend, this paper presents a human-centric AI framework that uses longhorizon forecasting for mitigation decision support in an integral pressurized water reactor (iPWR). The key idea is to convert predicted multivariate trajectories into a decision-relevant indicator, trip margin time (TMT), and to compare candidate mitigation actions against a no-action baseline on that basis. The forecasting engine is implemented as a two-stage latent prediction model consisting of an LSTM autoencoder and a latent predictor, where a short recent observation window is mapped to a compact latent representation and then reconstructed into a 2-h future trajectory. The framework is demonstrated through a case study of a pressurizer spray valve stuck-open transient, represented by the base malfunction RCSV09_7 and evaluated at two decision anchors, 35 s and 85 s. Candidate actions include valve manipulations, heater actions, and the no-action baseline. For each action assumption, future trajectories of trip-relevant variables are predicted, the earliest forecasted crossing of a triprelated setpoint is identified, and the corresponding TMT is computed. The case study shows that the proposed workflow clearly separates beneficial, near-neutral, and detrimental actions. In particular, while RCSV09_CLOSE provides a useful reference for margin recovery, the most meaningful actionable mitigation under the stuck-open assumption is RCSV08_OPEN, which substantially extends the available margin relative to no action. The results demonstrate that long-horizon forecasting can be operationalized as a human-centric mitigation advisory tool by transforming predicted trajectories into TMT-based action comparisons and prioritization. Future work will focus on procedure- and risk-based candidate-action generation, expansion to a broader set of abnormal iPWR scenarios, operator-facing HMI integration, and uncertainty-aware recommendation.

Accurate detection and timely restoration of faulty sensor signals are critical for ensuring safety and operational integrity in nuclear power plants, especially under accident conditions characterized by highly dynamic and nonlinear parameter behaviors. In this study, we propose a scenario-guided supervised autoencoder (SAE) framework that performs three integrated functions: (1) detecting faulty sensor signals during accident transients, (2) restoring corrupted signals to their unfaulted trajectories, and (3) preserving the quality of downstream accident diagnosis. Our framework employs a supervised variational autoencoder (VAE) with a long shortterm memory encoder, trained on accident scenario simulations from a compact nuclear simulator replicating key thermal-hydraulic behaviors of a pressurized water reactor. A classification decoder provides scenario-guided supervision, enabling the model to distinguish between fault-induced signal deviations and legitimate accident transient changes. Evaluation across nine accident scenarios and seven different sensor fault types shows that the proposed SAE achieves a fault detection true positive rate of 99.09% with 95.52% precision, compared to 95.50% and 91.99% for the conventional VAE. For signal restoration, 97.70% of restored signals fall within 15% mean absolute error of their unfaulted trajectories, compared to 69.50% for the VAE. The restoration-based approach recovers accident diagnosis accuracy to levels comparable to unfaulted conditions, outperforming fault isolation strategies for most fault types. Additional robustness analyses show that the proposed framework retains detection performance under additive Gaussian measurement noise up to sigma = 0.05 (5% of the normalized signal range) and achieves a 100% fault detection rate for up to three simultaneous sensor faults. These results suggest that scenario-guided supervised autoencoders can improve sensor signal integrity in safety-critical nuclear applications.

This study proposes a dynamic emergency operating procedural system to address the limitations of static, paper-based emergency operating procedures in nuclear power plants. Despite digital main control rooms, traditional procedures remain static and poorly suited to rapidly changing plant conditions. Operators must still search for relevant steps while ignoring those that do not apply, increasing workload and delaying decisions. The static format also hinders real-time tracking and verification, risking omission of safety-critical actions. To resolve these issues, this study developed the Emergency Guidance Intelligent System (EGIS), which provides real-time monitoring and required task blocks.EGIS comprises three core functions: the Task Block Browser, which delivers only necessary tasks in real-time using a functional-hierarchical task grouping framework; the Critical Safety Function Score Evaluator, which evaluates critical safety functions using fuzzy logic; and the Plant Status Monitor, which visualizes system status and consequence through Multilevel Flow Modeling. EGIS was verified using the Compact Nuclear Simulator with a Loss of Coolant Accident scenario. The results demonstrated that EGIS reduced operation time compared to conventional procedures while effectively replacing the role of traditional paper-based procedures. EGIS is expected to enhance the safety and efficiency of emergency operating procedures in nuclear power plants.

The 2011 Fukushima nuclear accident highlighted the need for thorough safety assessments of nuclear power plants, especially in the context of multi natural hazards that may occur simultaneously or sequentially. With climate change increasing both the frequency and intensity of natural hazards like typhoons, tsunamis, and extreme weather events, external event risk assessment frameworks have become essential. This study identifies natural hazards impacting nuclear power plants and introduces a multi external probabilistic safety assessment (PSA) framework. The proposed framework systematically evaluates nuclear plant risk under extreme and multi hazard using statistical data. The methodology incorporates steps such as hazard screening, correlation analysis, fragility analysis, accident scenario modeling, and risk analysis. Finally, a comprehensive risk assessment is conducted by combining results from the hazard and fragility analyses. A case study on nuclear power plants in Korea demonstrates the application and effectiveness of this multi external event PSA approach, aiming to enhance preparedness and resilience of nuclear facilities against increasingly multi natural hazards.

In the realm of dynamic probabilistic safety assessment, dynamic event trees (DETs) have been used to represent dynamic scenarios by branching at user-specified times or when an action is required by the operator and/or the systems to capture sequences of events based on their timing. While this approach provides a realistic and mechanistic analysis, it often results in an excessive number of branches, which reduces the explainability of the accident sequences and the visibility of the DET in comparison to traditional event trees based on representative scenarios. To address these challenges, this study proposes a method named DRAGON (Dynamic Risk Assessment through automatic accident sequence Generation using Optimized simulations for Nuclear power plants) for automatically generating accident sequences using optimized simulations for dynamic risk assessment. The proposed approach employs an optimized simulation algorithm to identify the limit surface between success and failure scenarios, which reduces the required simulations. Then a newly developed algorithm as a part of DRAGON provides the ability to analyze these simulations and control the complexity of the DET to present accident sequences in an interpretable form. To demonstrate the practicality and the effectiveness of the approach, two case studies on a loss of coolant accident were conducted, and comparisons with other data analysis methods were explored.

Dynamic probabilistic safety assessment (DPSA) can generate a large number of accident scenario sets whose analysis requires repeated execution of computationally expensive thermal-hydraulic (TH) system codes. Even when adaptive sampling reduces the number of scenarios requiring simulation, a substantial burden can remain because many sampled scenarios share identical accident timelines up to branching points. This paper proposes Optimized and Accelerated Simulation using Intermediate Storage (OASIS), a simulation-acceleration method that reuses such overlapping timelines. OASIS converts each scenario into a control-action timeline, identifies branching points, and stores checkpoint files generated by the restart capability of TH system codes. For a newly selected scenario, OASIS searches the checkpoint database for the furthest overlapping branching point and restarts the TH simulation from that checkpoint, thereby bypassing redundant calculations. A weight-based database management rule is also developed to retain checkpoints with high expected reuse benefits under finite storage budgets. The method was coupled with Deep-SAILS and applied to OPR1000 LOOP and cold-leg LOCA scenario sets using MAAP 5.05. In the 28,561-scenario LOOP case, Deep-SAILS selected 2,131 simulations requiring 24 h 29 min, whereas OASIS reduced the runtime to 8 h 51 min without memory constraints and 9 h 6 min with a 150-checkpoint limit. In the 6,875-scenario LOCA case, OASIS reduced the runtime from 11 h 45 min to 7 h 21 min, and to 8 h 27 min with a 40-checkpoint limit. The combined framework maintained high classification agreement with full-scenario results and remained effective for larger scenario sets.

This paper presents the monolithic three-dimensional (M3D) integration of a non-volatile out-of-plane MEMS memory directly above the complementary metal-oxide-semiconductor (CMOS) back end of line (BEOL) for reconfigurable logic applications. The integration process combines chemical–mechanical polishing (CMP) to planarize the BEOL surface with low-temperature electroplating to maintain full compatibility with the post-CMOS thermal budget. By fabricating the MEMS memory above the BEOL rather than constructing the device directly from BEOL metal layers, this approach overcomes the geometric and material constraints of prior in-plane BEOL-integrated MEMS memories and significantly reduces their typical k $\Omega $ –level on-resistance. The integrated see-saw MEMS memory demonstrates near-zero off-state leakage, retention exceeding $10^{6}$ seconds, and a low on-resistance of approximately $240~\Omega $ , with operation over 184 switchingcycles. These characteristics represent a substantial improvement over previous BEOL-integrated MEMS memories. Furthermore, a 2-input look-up table (LUT) implemented using four of these memory cells successfully performs AND, OR, and XOR logic operations by programming their non-volatile states, confirming on-chip reconfigurable logic capability. Overall, this work provides the first demonstration of an out-of-plane MEMS memory monolithically integrated on a CMOS platform and establishes a practical pathway toward CMOS-MEMS hybrid circuits. The results highlight the potential of such integration for future ultra-low-power, high-density, and reconfigurable logic systems.[2025-0229]

Diagnosis of abnormal events in nuclear power plants is important for ensuring operational safety and preventing human errors. Conventional deep learning-based diagnosis models often fail to ensure robust performance due to data discrepancies between simulator training data and actual plant data. To overcome data discrepancies, we enhanced the robustness of an abnormality diagnosis model through two key approaches. First, we developed a fuzzy-based feature extraction method to handle ambiguity and uncertainty between simulator and plant data by focusing on relatively long-term trends within time-series data rather than short-term data values. Second, we implemented a shallow-layer deep learning model to minimize overfitting on simulator data. Although deeper models may improve training accuracy, they do not always perform better in real-world applications. To evaluate our approach, we employed a synthetic plant dataset that reflects the discrepancies between simulator training data and actual plant data. The proposed model was trained solely on simulator data and tested on synthetic plant data. Experimental results show that the proposed model achieves a diagnostic accuracy of 99.7 %, while conventional models decline as data discrepancies increase. These outcomes illustrate the potential of our approach to improve the safety and reliability of nuclear power plant operations.

In nuclear power plants, operators can face cognitive workloads when diagnosing abnormal events due to the need to monitor numerous parameters and consider hundreds of potential scenarios. Artificial intelligence technologies have been proposed to support this process by providing diagnostic results; however, their lack of transparency can lead to out-of-the-loop unfamiliarity and distrust, hindering effective decision-making. To address these challenges, this study introduces a novel concept to enhance the understandability and trustworthiness of diagnostic support systems through Explainable Artificial Intelligence (XAI). The first method in the proposed concept rearranges monitoring parameters based on system structures to reflect parameter relationships. The second method refines explanations from XAI using Multilevel Flow Modeling (MFM) to ensure consistency with physical flow, and it visualizes diagnostic cause components on a plant map. By filtering out incomprehensible information and visualizing intuitive diagnostic causes, the system enables operators to identify expected causes of diagnostic results directly on the NPP map at the component or system level. This approach provides explainable and comprehensible support information, fostering trust in the system and improving diagnostic efficiency in abnormal situations.

This paper reports a complementary metal-oxide-semiconductor (CMOS) back-end-of-line (BEOL)compatible microelectromechanical non-volatile memory (MEM-NVM) with low on-resistance. The proposed seesaw design uses a torsional hinge with a low spring constant, enabling reliable memory retention through the stiction phenomenon. The fabricated device achieves a high on/off ratio of 10(6) in both '0' and '1' states with low on-resistance of 240 Omega. It demonstrates stable operation over 184 cycles and excellent data retention with less than 10% resistance variation after 10(6) seconds. Through the proposed design, we fabricated a MEM-NVM with high reliability and low on-resistance, positioning it as a breakthrough solution for next-generation CMOS-micro/nanoelectromechanical (M/NEMS) integrated circuits (ICs) for harsh environments and energy-efficient applications.

Nuclear power plants are equipped with various safety systems that reduce the frequency of accidents, but the risk of radiation release still exists in the event of an accident. In the event of a radiation release accident, residents near the power plant are evacuated. This study proposes evacuation strategies to minimize radiation exposure. It compares and analyzes the dispersion of radioactive materials and their impact on the population under typical weather conditions and natural disasters like typhoons. Under typical weather conditions, the Keyhole Strategy, based on the Puff model, is applied to establish real-time evacuation zones according to the dispersion area of radioactive materials, demonstrating effective evacuation and reduced radiation exposure. In contrast, during extreme weather conditions such as typhoons, changes make evacuation challenging. In such cases, a strategy of sheltering indoors until the typhoon passes, followed by evacuation, is more effective. This study highlights the potential of utilizing the precision of weather forecasting models and the Puff model to optimize real-time radiation impact assessments and evacuation strategies.

Dynamic probabilistic safety assessment (PSA) has been introduced due to the limitations of static-based PSA such as the difficulty to analyze dynamic sequences caused by stochastic random events. While various research has been performed to achieve this integration, quantifying risk in dynamic PSA is still challenging because operator response models that can provide a branch probability according to the timing of operator action in dynamic scenarios have not yet been addressed. Existing human reliability analysis (HRA) models only consider the time given to operators for actions insofar as it can impact the failure probabilities of the human actions, despite the timing of the actions being a vital element of HRA for dynamic scenarios. This paper proposes an operator action timing-based human reliability evaluation method for dynamic PSA to evaluate the distribution of operator action timing. The method covers operator action timings with a model that convolutes two time distribution functions to provide the probability of the success or failure of an operator action. To demonstrate the practicality of the proposed method and its effectiveness, a case study and uncertainty analysis for a small break loss of coolant accident with two operator tasks were conducted.

This paper introduces the coupling methodology of the thermo-mechanical code, FRAPTRAN/CUPID, and discusses the verification and validation calculations of the coupled FRAPTRAN/CUPID. In the verification calculation, which utilized 3 x 3 fuel rod channels, the coupled calculation produced consistent results compared to the standalone CUPID calculation. The validation calculations, based on two international and domestic LBLOCA tests (OECD-Halden IFA-650.5 and ICARUS-RT-20-02), demonstrate that the coupled FRAPTRAN/CUPID code effectively simulates the thermal-hydraulic behaviors of nuclear fuel rods during LBLOCA. It also provides valuable thermo-mechanical information regarding the deformation and burst of the nuclear fuel rods during LBLOCA. Finally, the paper presents an OPR1000 SLB calculation as a reactor application, utilizing the FRAPTRAN/MASTER/CUPID/MARS framework.

Large-scale infrastructures, such as chemical plants and nuclear power plants (NPPs), are pivotal for modern civilization as they provide vital resources and energy. However, their operation introduces significant risks, as demonstrated by the tragic accidents at Bhopal and Fukushima. While extensive research has been conducted to improve the safety of these safety–critical systems, the human factor remains as a significant concern. In recent years, as artificial intelligence (AI) is being widely adopted in various fields, AI may be a solution for supporting operators and, ultimately, for reducing the overall risk of safety–critical systems such nuclear and chemical plants. This review discusses the application of AI in NPP operations, with a focus on event diagnosis, signal validation, prediction, and autonomous control. Various application examples are presented, highlighting the limitations of classical approaches and the potential for AI overcome such limitations to enhance the safety and efficiency of NPP operations. This work is expected to stimulate further investigation into the application of AI to support operators in not only NPPs but also other safety–critical systems, such as chemical plants.

An issue regarding the incorporation of software reliability within the nuclear power plant (NPP) probabilistic risk assessment model has emerged in the licensing processes of digitalized NPPs. Since software failure induces common-cause failure of the processor modules, the reliability of the software used in the NPP safety-critical instrumentation and control systems must be quantified and verified with proper test cases and environments.In this study, a software testing method based on the minimal cut set (MCS)-based exhaustive test case generation scheme is proposed where the software logic model is developed from available information on the software development and the MCSs that represent the necessary and sufficient conditions for the software variables' states to produce safety software outputs are generated. The MCSs are then converted into the test cases, which can be used as inputs to the test bed to verify that the test cases produce correct outputs after software execution. The effectiveness of the proposed method is demonstrated with the safety-critical trip logic software of the APR-1400 reactor protection system. The method provides a systematic way to conduct exhaustive software testing and prove the functionality of the nuclear safety software based on the test result without uncertainties.

Carried out was the assessment of the radiation safety for the recycling workers of the radioactive concrete generated from nuclear power plants and nearby residents. Decommissioning of a nuclear power plant with an electric output of 1000 MW generates approximately 500,000 tons of concrete waste. It is important to reduce the volume of concrete waste to reduce the costs of decommissioning projects. For this purpose, the low-level radioactive cement powder from bioshield concrete is recycled as a solidifying agent. The number of radioactive concrete waste drums to be disposed is reduced by extracting cement from the radioactive concrete and using it as a solidifying agent, rather than using cement from general industrial waste. Various exposure scenarios exist, and it is necessary to evaluate the radiation dose in each scenario. In this study, seven scenarios were selected and evaluated. The radioactive exposure to facility workers, nuclear power plant workers, and landfill management workers is within the annual dose limit, even if they work 2000 h per year. A loader can work close to the drum for 90-93 h per year, and a truck driver can drive 621 one-way trucks per year. Dose evaluations for road users and residents after landfill closure showed that both had values far below the dose limit for the public. Bioshield concrete recycling was confirmed to be radiologically safe in all scenarios if certain conditions were satisfied. Thus, this study showed that cement powder from bioshield concrete could be used safely without exceeding the radiation safety limits for workers and the public when recycling concrete.