In this paper, we investigate the feasibility of a strategy of fault detection capable of controlling misclassification probabilities, i.e., balancing false and missed alarms. The novelty of the ...proposed strategy consists of i) a signal grouping technique and signal reconstruction modeling technique (one model for each subgroup), and ii) a statistical method for defining the fault alarm level. We consider a real case study concerning 46 signals of the Reactor Coolant Pump (RCP) of a typical Pressurized Water Reactor (PWR). In the application, the reconstructions are provided by a set of Auto-Associative Kernel Regression (AAKR) models, whose input signals have been selected by a hybrid approach based on Correlation Analysis (CA) and Genetic Algorithm (GA) for the identification of the groups. Sequential Probability Ratio Test (SPRT) is used to define the alarm level for a given expected classification performance. A practical guideline is provided for optimally setting the SPRT parameters' values.
In this paper, an ensemble approach is proposed for prediction of time series data based on a Support Vector Regression (SVR) algorithm with RBF loss function. We propose a strategy to build diverse ...sub-models of the ensemble based on the Feature Vector Selection (FVS) method of Baudat & Anouar (2003), which decreases the computational burden and keeps the generalization performance of the model. A simple but effective strategy is used to calculate the weights of each data point for different sub-models built with RBF-SVR. A real case study on a nuclear power production component is presented. Comparisons with results given by the best single SVR model and a fixed-weights ensemble prove the robustness and accuracy of the proposed ensemble approach.
The objective of the present work is to develop a novel approach for combining in an ensemble multiple base clusterings of operational transients of industrial equipment, when the number of clusters ...in the final consensus clustering is unknown. A measure of pairwise similarity is used to quantify the co-association matrix that describes the similarity among the different base clusterings. Then, a Spectral Clustering technique of literature, embedding the unsupervised K-Means algorithm, is applied to the coassociation matrix for finding the optimum number of clusters of the final consensus clustering, based on Silhouette validity index calculation. The proposed approach is developed with reference to an artificial casestudy, properly designed to mimic the signal trend behavior of a Nuclear Power Plant (NPP) turbine during shut-down. The results of the artificial case have been compared with those achieved by a state-of-art approach, known as Clusterbased Similarity Partitioning and Serial Graph Partitioning and Fill-reducing Matrix Ordering Algorithms (CSPAMETIS). The comparison shows that the proposed approach is able to identify a final consensus clustering that classifies the transients with better accuracy and robustness compared to the CSPA-METIS approach. The approach is, then, validated on an industrial case concerning 149 shut-down transients of a NPP turbine.
In this paper, a novel Dynamic-Weighted Probabilistic Support Vector Regression-based Ensemble (DW-PSVR-ensemble) approach is proposed for prognostics of time series data monitored on components of ...complex power systems. The novelty of the proposed approach consists in (i) the introduction of a signal reconstruction and grouping technique suited for time series data, (ii) the use of a modified Radial Basis Function (RBF) kernel for multiple time series data sets, (iii) a dynamic calculation of sub-models weights for the ensemble, and (iv) an aggregation method for uncertainty estimation. The dynamic weighting is introduced in the calculation of the sub-models' weights for each input vector, based on Fuzzy Similarity Analysis (FSA). We consider a real case study involving 20 failure scenarios of a component of the Reactor Coolant Pump (RCP) of a typical nuclear Pressurized Water Reactor (PWR). Prediction results are given with the associated uncertainty quantification, under the assumption of a Gaussian distribution for the predicted value.
Framework for reconciliating the consensus clusterings for a fleet of P components.
Display omitted
•A framework for incremental learning of clustering results of a fleet is proposed.•Different ...unsupervised clusters for each individual plant are obtained.•The framework is applied on the reconciliation of the turbine transients of two NPPs.•Results show that the proposed approach is superior to other alternative approaches.
When a fleet of similar Systems, Structures and Components (SSCs) is available, the use of all the available information collected on the different SSCs is expected to be beneficial for the diagnosis purpose. Although different SSCs experience different behaviours in different environmental and operational conditions, they maybe informative for the other (even if different) SSCs. In the present work, the objective is to build a fault diagnostic tool aimed at capitalizing the available data (vibration, environmental and operational conditions) and knowledge of a heterogeneous fleet of P Nuclear Power Plants (NPPs) turbines. To this aim, a framework for incrementally learning different clusterings independently obtained for the individual turbines is here proposed. The basic idea is to reconciliate the most similar clusters across the different plants. The data of shut-down transients acquired from the past operation of the P NPPs turbines are summarized into a final, reconciliated consensus clustering of the turbines behaviors under different environmental and operational conditions. Eventually, one can distinguish, among the groups, those of anomalous behavior and relate them to specific root causes. The proposed framework is applied on the shut-down transients of two different NPPs. Three alternative approaches for learning data are applied to the case study and their results are compared to those obtained by the proposed framework: results show that the proposed approach is superior to the other approaches with respect to the goodness of the final consensus clustering, computational demand, data requirements, and fault diagnosis effectiveness.
Empirical methods for fault diagnosis usually entail a process of supervised training based on a set of examples of signal evolutions “labeled” with the corresponding, known classes of fault. ...However, in practice, the signals collected during plant operation may be, very often, “unlabeled”, i.e., the information on the corresponding type of occurred fault is not available. To cope with this practical situation, in this paper we develop a methodology for the identification of transient signals showing similar characteristics, under the conjecture that operational/faulty transient conditions of the same type lead to similar behavior in the measured signals evolution. The methodology is founded on a feature extraction procedure, which feeds a spectral clustering technique, embedding the unsupervised fuzzy C-means (FCM) algorithm, which evaluates the functional similarity among the different operational/faulty transients. A procedure for validating the plausibility of the obtained clusters is also propounded based on physical considerations. The methodology is applied to a real industrial case, on the basis of 148 shut-down transients of a Nuclear Power Plant (NPP) steam turbine.
•Operational/faulty transient conditions of the same type lead to similar transients.•We present a methodology for the identification of similarities among transients.•A feature extraction procedure feeds a spectral clustering technique.•Spectral clustering embeds an unsupervised fuzzy C-means (FCM) algorithm.•The methodology is applied to 148 shut-down transients of a Nuclear Power Plant (NPP).
► A modified Probabilistic Support Vector Regression (PSVR) method is used in condition monitoring of nuclear power plant. ► Novel and effective hyperparameter tuning and model identification methods ...are proposed. ► The PSVR-based procedure is effectively used for short-term forecasting of out-of-control parameters. ► A comparison is performed between PSVR and other prediction methods, proving the effectiveness of the former approach.
In this paper, an approach for the prediction of the condition of Nuclear Power Plant (NPP) components is proposed, for the purposes of condition monitoring. It builds on a modified version of the Probabilistic Support Vector Regression (PSVR) method, which is based on the Bayesian probabilistic paradigm with a Gaussian prior. Specific techniques are introduced for the tuning of the PSVR hyerparameters, the model identification and the uncertainty analysis. A real case study is considered, regarding the prediction of a drifting process parameter of a NPP component.
The digitalization of nuclear power plants, with the rapid growth of information technology, opens the door to the development of new methods of condition-based maintenance. In this work, a ...semi-supervised method for characterizing the level of degradation of nuclear power plant components using measurements collected during plant operational transients is proposed. It is based on the fusion of selected features extracted from the monitored signals. Feature selection is formulated as a multi-objective optimization problem. The objectives are the maximization of the feature monotonicity and trendability, and the maximization of a novel measure of correlation between the feature values and the results of non-destructive tests performed to assess the component degradation. The features of the Pareto optimal set are normalized and the component degradation level is defined as the median of the obtained values. The developed method is applied to real data collected from steam generators of pressurized water reactors. It is shown able to identify degradation level with errors comparable to those obtained by ad-hoc non-destructive tests.
Display omitted
•A semi-supervised method for identifying the degradation level of industrial components is proposed.•The method is based on the three steps of feature extraction, evaluation and selection.•The extracted features performance is evaluated considering monotonicity, trendability and a novel correlability index.•The method is applied to fleets of steam generators in nuclear power plants.•The method identifies the component degradation level with performance comparable to those of non-destructive tests.