• Energy Research

Abstract The security and stability of systems and components in operating nuclear power plants (NPPs) are extremely important factors that determine the plants reliability and service life. An essential system that help reduce the maintenance costs and improve the reliability of NPPs is an efficient fault diagnosis system. In this paper, an optimized fault diagnosis framework is proposed to efficiently identify system-level failures and their severities to guarantee the sustainability of NPPs. To facilitate the recognition of the real-time failure types and to extract both the constraint relationships and fault regularities of system-level parameters, the least squares support vector machine (LS-SVM) method was introduced at the fault diagnosis step. The severity assessment to estimate the degree of failures was subsequently performed using a method derived from gaussian process regression (GPR). To overcome the challenge of selecting hyperparameters of GPR, the particle swarm optimization (PSO) was applied to search for the optimal hyperparameters of GPR. The PSO intelligent search strategy implemented to obtain a fault severity assessment model ultimately aided operators to make an informed decision on the operating conditions of the plant. Simulations carried out based on the proposed fault diagnosis framework demonstrate the accuracy and reliability of the methodology, as well as the availability of support for the stable operation of NPPs. This proposed comprehensive fault diagnosis framework is suitable for multi-dimensional monitoring of NPPs.

Abstract The security and stability of systems and components in operating nuclear power plants (NPPs) are extremely important factors that determine the plants reliability and service life. An essential system that help reduce the maintenance costs and improve the reliability of NPPs is an efficient fault diagnosis system. In this paper, an optimized fault diagnosis framework is proposed to efficiently identify system-level failures and their severities to guarantee the sustainability of NPPs. To facilitate the recognition of the real-time failure types and to extract both the constraint relationships and fault regularities of system-level parameters, the least squares support vector machine (LS-SVM) method was introduced at the fault diagnosis step. The severity assessment to estimate the degree of failures was subsequently performed using a method derived from gaussian process regression (GPR). To overcome the challenge of selecting hyperparameters of GPR, the particle swarm optimization (PSO) was applied to search for the optimal hyperparameters of GPR. The PSO intelligent search strategy implemented to obtain a fault severity assessment model ultimately aided operators to make an informed decision on the operating conditions of the plant. Simulations carried out based on the proposed fault diagnosis framework demonstrate the accuracy and reliability of the methodology, as well as the availability of support for the stable operation of NPPs. This proposed comprehensive fault diagnosis framework is suitable for multi-dimensional monitoring of NPPs.

Abstract The complexity and safety requirements for Nuclear power plants (NPP) warrant a reliable fault diagnosis approach. In this paper, we present a fault diagnosis method based on Correlation Analysis and Deep Belief Network. We utilized the feature selection capability of Correlation Analysis for dimensionality reduction and deep belief network for fault identification. We also discussed the implementation of the algorithm and the process of model building that is characteristics of NPP. To illustrate the performance of the proposed fault diagnosis model, we utilized Personal Computer Transient Analyzer (PCTRAN). In addition, we also compared the fault diagnostic results from back-propagation neural network and support vector machine with our method. The results show that the proposed method has obvious advantages over other methods, and would be of profound guiding significance to the fault diagnosis of NPP.

Abstract The complexity and safety requirements for Nuclear power plants (NPP) warrant a reliable fault diagnosis approach. In this paper, we present a fault diagnosis method based on Correlation Analysis and Deep Belief Network. We utilized the feature selection capability of Correlation Analysis for dimensionality reduction and deep belief network for fault identification. We also discussed the implementation of the algorithm and the process of model building that is characteristics of NPP. To illustrate the performance of the proposed fault diagnosis model, we utilized Personal Computer Transient Analyzer (PCTRAN). In addition, we also compared the fault diagnostic results from back-propagation neural network and support vector machine with our method. The results show that the proposed method has obvious advantages over other methods, and would be of profound guiding significance to the fault diagnosis of NPP.

Abstract The principal component analysis (PCA) method has been widely used in sensor fault detection. However, outliers of training data may affect the projection directions of both principal component (PC) and residual space, thereby reducing the fault detection rate (FDR). The high sensitivity of PCA to random noise in the test data can also lead to an increase in the false alarm rate (FAR). To improve the performance of the PCA, this paper proposes a robust PCA approach for sensor fault detection in nuclear power plants (NPPs). A statistical method based on Euclidean distance is used to clean outliers in the training data pre-processing phase. Subsequently in the fault detection phase, the moving average (MA) filtering method is adopted to process Q-statistic to reduce false alarms caused by random noise in the test data. Simulation and plant signals are used to verify the effectiveness of the proposed method. Finally, comparisons with the conventional PCA, auto-associative kernel regression (AAKR) and multivariate state estimation technique (MSET) highlight the advantages of the proposed method.

Abstract The principal component analysis (PCA) method has been widely used in sensor fault detection. However, outliers of training data may affect the projection directions of both principal component (PC) and residual space, thereby reducing the fault detection rate (FDR). The high sensitivity of PCA to random noise in the test data can also lead to an increase in the false alarm rate (FAR). To improve the performance of the PCA, this paper proposes a robust PCA approach for sensor fault detection in nuclear power plants (NPPs). A statistical method based on Euclidean distance is used to clean outliers in the training data pre-processing phase. Subsequently in the fault detection phase, the moving average (MA) filtering method is adopted to process Q-statistic to reduce false alarms caused by random noise in the test data. Simulation and plant signals are used to verify the effectiveness of the proposed method. Finally, comparisons with the conventional PCA, auto-associative kernel regression (AAKR) and multivariate state estimation technique (MSET) highlight the advantages of the proposed method.

Abstract Nuclear power plant is a highly safety required system which has multi- operating condition in different power mode, and it requires a more advanced technology to realize condition monitoring. To improve the condition monitoring techniques, a mixed condition monitoring method based on sparse autoencoder and isolation forest is proposed to realize the condition monitoring of nuclear power plant, where sparse autoencoder is responsible for data feature extraction and dimensionality reduction, and isolation forest is responsible for the anomaly monitoring of nuclear power plant. The proposed method can transform high-dimensional data into a low-dimensional space, remove the redundancy of the data, and then identify the state through a high-performance monitoring model, thereby improving monitoring efficiency and accuracy. In order to expound the performance of the condition monitoring model proposed in this paper, we select one operating condition and two operating conditions for testing. We also obtained the condition monitoring results of local outlier factor and one-class support vector machine to compare with our method. From the results, it can be known that sparse autoencoder can extract the nature of operating data, and monitoring accuracy of 100% and 98% can be achieved under one operating condition and two operating conditions by isolation forest method, respectively. Compared with other methods, the proposed method has obvious advantages. This research has important implications for the condition monitoring of nuclear power plant and the system with multi-operating conditions.

Abstract Nuclear power plant is a highly safety required system which has multi- operating condition in different power mode, and it requires a more advanced technology to realize condition monitoring. To improve the condition monitoring techniques, a mixed condition monitoring method based on sparse autoencoder and isolation forest is proposed to realize the condition monitoring of nuclear power plant, where sparse autoencoder is responsible for data feature extraction and dimensionality reduction, and isolation forest is responsible for the anomaly monitoring of nuclear power plant. The proposed method can transform high-dimensional data into a low-dimensional space, remove the redundancy of the data, and then identify the state through a high-performance monitoring model, thereby improving monitoring efficiency and accuracy. In order to expound the performance of the condition monitoring model proposed in this paper, we select one operating condition and two operating conditions for testing. We also obtained the condition monitoring results of local outlier factor and one-class support vector machine to compare with our method. From the results, it can be known that sparse autoencoder can extract the nature of operating data, and monitoring accuracy of 100% and 98% can be achieved under one operating condition and two operating conditions by isolation forest method, respectively. Compared with other methods, the proposed method has obvious advantages. This research has important implications for the condition monitoring of nuclear power plant and the system with multi-operating conditions.

Abstract Extracting features for early failure detection in rotating machinery of nuclear power plants (NPPs) is difficult because in the early stages of failure the impact on the vibration signals is weak. To improve early fault detection in rotating machinery, a fault feature extraction method based on the combination of parameter-adaptive Variational Mode Decomposition (VMD) and Teager energy operator (TEO) is proposed in this paper. Firstly, we introduce the maximum weighted kurtosis index (WKI) as the objective function, and the Artificial Bee Colony (ABC) is used to optimize the VMD parameters. Then, the optimized VMD is used to decompose the vibration signal into multiple intrinsic mode functions (IMFs). Finally, TEO is used to demodulate the sensitive mode with the maximum WKI and identify the fault frequencies. Simulation and experiment show that the early fault features in vibration signals can be effectively extracted by the proposed method, and the comparisons with other three methods highlight the advantages of the proposed method.