Unsupervised anomaly detection in pressurized water reactor digital twins using autoencoder neural networks

The LOCA analysis for the advanced pressurized water reactor (PWR) is very important and the methods on it are developing. There are two basic approaches for LOCA (loss of coolant accident) licensing at current. One is based on the conservative requirement of Appendix K of 10CFR50.46 of USNRC, and another is the best estimate (BE) analysis methodology which needs strict sensitivity and uncertainty analysis. The results achieved by the best estimate analysis are closer to the reality than those achieved by the conservative methodology, and the realistic BELOCA analysis in nuclear realm becomes an international trend currently although its development still meet lots of challenges. The research and design on AP1000 to be built in China and larger advanced pressurized water reactor (CAP1400 or CAP1700) as one of Chinese national science & technology major project is in progress. The reliable licensing LOCA analysis as one of the most important accident safety analysis is absolutely necessary. There are three ways to get the code applied in licensing accident analysis: the first way is developing code based on the best estimated methodology with strict uncertainty analysis, the second way is to develop new analysis code based on the conservative Appendix K, and the third way is improving the current system analysis code, which had been verified and validated by many cases, to satisfy the requirements of Appendix K. The last one may be the most feasible way for the AP1000 design with high efficiency and economic competition. Some code like RELAP5 has been used for LOCA analysis, and its results showed good agreement with the test data. RELAP is the transient thermal-hydraulic system analysis code developed by Idaho National Laboratory, in which some model and correlations are not consistent with the conservative requirements of Appendix K, so it can not be applied for licensing LOCA analysis and evaluation directly. In this paper the way to develop analysis code for LOCA license is discussed, and some areas in RELAP code needed to be modified for according with Appendix K are also described, which will be helpful for the advanced PWR design and development in China.

A novel damage detection approach based on Auto-encoder neural network is proposed to identify damage in beam-like bridges subjected to a moving mass. In this approach, several sensors are used to measure structural vibration responses during a mass moving across the bridge. An auto-encoder (AE) neural network is designed to extract features from the measured responses. A fixed moving window is used to cut out the time-domain responses to generate inputs of the AE neural network. Moreover, some constraints are applied on the hidden layer to improve the performance of the AE network in training process. When the training is complete, the encoder was regarded as a feature extractor. And the damage index is defined as the cosine distance between two feature vectors obtained from adjacent data windows. By moving the window along the measured vibration data, we can calculate a damage index series and locate the damage position of the structure. To demonstrate the performance of the proposed method, numerical simulation is carried out. The results show that the proposed method can accurately locate both single and multiple damages using acceleration response. It infers the proposed method is promising for structural damage detection.

Tiny machine learning (TinyML) is a fast-growing research area committed to democratizing deep learning for all-pervasive microcontrollers (MCUs). Challenged by the constraints on power, memory, and computation, TinyML has achieved significant advancement in the last few years. However, the current TinyML solutions are based on batch/offline setting and support only the neural network's inference on MCUs. The neural network is first trained using a large amount of pre-collected data on a powerful machine and then flashed to MCUs. This results in a static model, hard to adapt to new data, and impossible to adjust for different scenarios, which impedes the flexibility of the Internet of Things (IoT). To address these problems, we propose a novel system called TinyOL (TinyML with Online-Learning), which enables incremental on-device training on streaming data. TinyOL is based on the concept of online learning and is suitable for constrained IoT devices. We experiment TinyOL under supervised and unsupervised setups using an autoencoder neural network. Finally, we report the performance of the proposed solution and show its effectiveness and feasibility.

Failure probabilities of SiC clad fuel during a LOCA in public acceptable simple SMR (PASS)

The rapid advancement of Industry 4.0 and the Internet of Things (IoT) has underscored the importance of effective diagnostics for electric motors, which are critical components in industrial, energy, and transportation systems. Traditional diagnostic methods, such as vibration, thermographic, and electromagnetic analyses, often lack adaptability and pre-dictive capabilities, limiting their ability to proactively identify faults under variable operating conditions. The digital twin concept, a virtual representation of physical assets synchronized with real-time sensor data, offers a transformative approach to technical condition monitoring. This paper reviews the state of diagnostic methods for electric mo-tors and evaluates the integration of digital twins to enhance diagnostic accuracy and pre-dictive maintenance. By leveraging platforms like Ansys Twin Builder and MATLAB/Simulink, digital twins enable real-time simulation, fault detection, and opera-tional optimization. The study proposes a novel methodology for diagnosing electric motor conditions using digital twins, incorporating mathematical modeling and real-time signal processing. Key findings include improved fault detection accuracy (MAE < 2.5%, RMSE < 3.1%) and the ability to simulate typical defects like rotor imbalance and insula-tion breakdown. The results demonstrate the potential of digital twins to revolutionize condition-based and predictive maintenance, offering scalable solutions for industrial applications. Future prospects include integrating machine learning for adaptive diagnos-tics and expanding digital twin applications to complex electromechanical systems.

Imbalanced data classification problem has always been a hotspot in the field of machine learning research. Pointing to the overfitting and noise problems of oversampling algorithm when synthesizing new minority class samples, the current study proposed a stacked denoising autoencoder neural network (SDAE) algorithm based on cost-sensitive oversampling, combining the cost-sensitive learning with denoising autoencoder neural network. The proposed algorithm can not only oversample minority class sample through misclassification cost, but it can denoise and classify the sampled dataset. Experiment shows that, compared with the traditional stacked autoencoder neural network (SAE) and oversampling autoencoder neural network without denoising process (OS-SAE), the proposed algorithm improves the classification accuracy of minority class of imbalanced datasets.

Combining autoencoder neural network and Bayesian inversion to estimate heterogeneous permeability distributions in enhanced geothermal reservoir: Model development and verification

Being able to give real time on-board advice, without depending on extensive sets of measured data, is the ultimate goal of the digital twin concept. Ideally, the models used in a digital twin only rely on current in-service data, although they have been built using simulated and possibly some measured data. Working with just the 6-DOF motions of a ship, can the local sea state reliably be estimated using the digital twin concept? Does a general model exist to do so, without the need to measure or simulate the particular ship? In this paper, we discuss how simulations of an advancing ship, subjected to various sea states, can be used to estimate the relative wave direction from in-service motion measurements of the corresponding ship. Various types of neural networks are used and evaluated with simulated data and measured data. In order to study the generalization power of the neural networks, a range of ships has been simulated, with varying lengths, drafts and geometries. Neural networks have been trained on selections of the ships in this extended training set and evaluated on the remaining ships. Results show that the developed neural networks give a remarkable performance in simulation data. Furthermore, generalization over geometry is very good, opening the door to train a general model for estimating sea state characteristics. Using the same model for in-service measurements does not perform well enough yet and further research is required. The paper will include discussion on possible causes for this performance gap and some promising ideas for future work.

This paper proposes a state estimation method based on the digital twin (DT) concept. With the increasing complexity of the power grid, the state estimation is more needed to maintain the stability of operation. The availability of digital twin model can assist in estimating the status of the power grid. Based on the measured data from the phasor measurement unit (PMU), the DT can monitor the power grid condition. Additionally, it can estimate the possible states in the power grid with the future event. In this paper, the state estimation of the inverter dominated grid by DT model is formulated. To build the DT model, the neural network (NN) is utilized to emulate the dynamic feature of the inverter. The training data of NN in DT is obtained from the reference model. The reference model is created based on traditional dynamic state equations of inverter and power grid. To validate the accuracy of DT model, the simulation results of the reference model are firstly compared with the results of the DT model after the training process. Afterwards, two additional scenarios are simulated initially by DT based on the Cigre benchmark grid to validate the accuracy of state estimation in the power grid. The results illustrate that the state estimation by DT has high similarity degree with the reference model.

Chapter 3 - Five-Dimension Digital Twin Modeling and Its Key Technologies

Presented are experiment results of the Large Scale Test Facility (LSTF) conducted at the Japan Atomic Energy Agency (JAEA) with a focus on core exit thermocouple (CET) performance to detect core overheat during a vessel top break loss-of-coolant accident (LOCA) simulation experiment. The CET temperatures are used to start accident management (AM) action to quickly depressurize steam generator (SG) secondary sides in case of core temperature excursion. Test 6-1 is the first test of the OECD/NEA ROSA Project started in 2005, simulating withdraw of a control rod drive mechanism penetration nozzle at the vessel top head. The break size is equivalent to 1.9% cold leg break. The AM action was initiated when CET temperature rose up to 623K. There was no reflux water fallback onto the CETs during the core heat-up period. The core overheat, however, was detected with a time delay of about 230s. In addition, a large temperature discrepancy was observed between the CETs and the hottest core region. This paper clarifies the reasons of time delay and temperature discrepancy between the CETs and heated core during boil-off including three-dimensional steam flows in the core and core exit. The paper discusses applicability of the LSTF CET performance to pressurized water reactor (PWR) conditions and a possibility of alternative indicators for earlier AM action than in Test 6-1 is studied by using symptom-based plant parameters such as a reactor vessel water level detection.

Behavior of Emergency Core Cooling System (ECCS) during the early stage of Loss of Coolant Accident (LOCA) for APR 1400 with Flownex software

Migrating the industry and society into digital area have gradually been increased in recent years. Digitalization has made our life more efficient and faster. As a result of digitalization, the Fourth Industrial Revolution, also known as Industry 4.0, was born. The Fourth Industrial Revolution, like previous revolutions, aims to facilitate life and improve the quality of life. Two terms are especially important for digitalization. These terms are Industry 4.0 and Internet of Things (IoT). Industry 4.0 and the IoT focus on the networking and automation of devices both used human being and industry. Digitalization and IoT technologies play an important role in improving energy efficiency. Digital twin (DT) technology is one of the most important key of Industry 4.0 technologies under digitalization. DT represents a digital copy of a physical system or entity. Physical objects and DT models created interact with each other. DT technologies have a wide range from manufacturing to healthcare sectors and from maritime transportation to the heavy industry. DT provides many conveniences for industries to grow with digitalization. Meanwhile, DT draws attention with its development in the academic field. In this study, an overview of DT is presented, a DT roadmap is proposed, some algorithms emerging in the concept of DT for industrial processes are explained and a case study is analyzed. This roadmap shows different methods to be applied in order to make accurate forecasting for the future applications and sustainable manufacturing with DT. As a result, DT concept is thought to attract much attention in the near future.

<div class="section abstract"><div class="htmlview paragraph">To facilitate the construction of a robust transport infrastructure, it is essential to implement a digital transformation of the current highway system. The concept of digital twins, which are virtual replicas of physical assets, offers a novel approach to enhancing the operational efficiency and predictive maintenance capabilities of highway networks. The present study begins with an exhaustive examination of the demand for the smart highway digital twin model, underscoring the necessity for a comprehensive framework that addresses the multifaceted aspects of digital transformation. The framework, as proposed, is composed of six integral components: spatiotemporal data acquisition and processing, multidimensional model development, model integration, application layer construction, model iteration, and model governance. Each element is critical in ensuring the fidelity and utility of the digital twin, which must accurately reflect the dynamic nature of highway systems. The methodology for constructing a smart highway digital twin model is explored through a systematic approach that encompasses three pivotal stages. The first stage involves the comprehensive perception of spatiotemporal data, the foundation for any digital twin. The second stage pertains to entity modeling, where the physical assets of the highway system are digitized, thus creating a virtual representation that can be manipulated and analyzed. The final stage is real-time state modeling, which enables the digital twin to simulate the current state of the highway system, thereby providing real-time feedback and predictive analytics. This work aims to contribute to the theoretical and technical discourse surrounding smart highway digital twins, offering insights that can inform the development and practical application of such models. By adhering to the proposed framework and methodology, workers in the transportation sector can leverage the potential of digital twins to enhance safety, efficiency, and sustainability within the highway infrastructure ecosystem.</div></div>

The concept of Digital Twins is still in an infant state of development. Digital Twins are often built as a tool to aid in better understanding of physical systems through simulation. They can be used to visualize information during operations and provide instructions during training or execution of procedures. The use of Digital Twins to test is appealing as it can be done quickly and safely. However, testing without inclusion of the physical system can lead to a reality gap. The reality gap can lead to high risks when applying concepts tested on digital Twins to the physical system directly. Sometimes interaction with the physical system is unfeasible. In this paper, we present an experimental laboratory that we built to provide a platform for the development of high quality Digital Twins through a feedback loop. The physical system is a Palfinger crane. Our replicate physical twin is a Universal collaborative industrial robot model UR16e due to its similar anatomy to the crane. The RoboDK simulation software was used to rapidly develop a digital twin of the UR16e. We demonstrate a solution to the interoperability problem in digital Twins using the monitoring adaptation loop from the Autonomic Adaptation System of the Arrowhead Framework.