Roozbeh Razavi-Far

Model-based fault detection and isolation of a steam generator using neuro-fuzzy networks

This paper presents a neuro-fuzzy (NF) networks based scheme for fault detection and isolation (FDI) of a U-tube steam generator (UTSG) in a nuclear power plant. Two types of NF networks are used. A NF based learning and adaptation of Takagi-Sugeno (TS) fuzzy models is used for residual generation, while for residual evaluation a NF network for Mamdani models is used. The NF network for Takagi-Sugeno models is trained with data collected from a full scale UTSG simulator and is used for generating residuals in the fault detection step. A locally linear neuro-fuzzy (LLNF) model is used in the identification of the steam generator. This model is trained using the locally linear model tree (LOLIMOT) algorithm. In the fault isolation part, genetic algorithms are employed to train a Mamdani type NF network, which is used to classify the residuals and take the appropriate decision regarding the actual behavior of the process. Furthermore, a qualitative description of faults is then extracted from the fuzzy rules obtained from the Mamdani NF network. Experimental results presented in the final part of the paper confirm the effectiveness of this approach.

Classifier-ensemble incremental-learning procedure for nuclear transient identification at different operational conditions

An important requirement for the practical implementation of empirical diagnostic systems is the capability of classifying transients in all plant operational conditions. The present paper proposes an approach based on an ensemble of classifiers for incrementally learning transients under different operational conditions. New classifiers are added to the ensemble where transients occurring in new operational conditions are not satisfactorily classified. The construction of the ensemble is made by bagging; the base classifier is a supervised Fuzzy C Means (FCM) classifier whose outcomes are combined by majority voting. The incremental learning procedure is applied to the identification of simulated transients in the feedwater system of a Boiling Water Reactor (BWR) under different reactor power levels.

Dynamic Weighting Ensembles for Incremental Learning and Diagnosing New Concept Class Faults in Nuclear Power Systems

Key requirements for the practical implementation of empirical diagnostic systems are the capabilities of incremental learning of new information that becomes available, detecting novel concept classes and diagnosing unknown faults in dynamic applications. In this paper, a dynamic weighting ensembles algorithm, called Learn++.NC, is adopted for fault diagnosis. The algorithm is specially designed for efficient incremental learning of multiple new concept classes and is based on the dynamically weighted consult and vote (DW-CAV) mechanism to combine the classifiers of the ensemble. The detection of unseen classes in subsequent data is based on thresholding the normalized weighted average of outputs (NWAO) of the base classifiers in the ensemble. The detected unknown classes are classified as unlabeled until their correct labels can be assigned. The proposed diagnostic system is applied to the identification of simulated faults in the feedwater system of a boiling water reactor (BWR).

Efficient residuals pre-processing for diagnosing multi-class faults in a doubly fed induction generator, under missing data scenarios

Abstract This paper focuses on the development of a pre-processing module to generate the latent residuals for sensor fault diagnosis in a doubly fed induction generator of a wind turbine. The pre-processing module bridges a gap between the residual generation and decision modules. The inputs of the pre-processing module are batches of residuals generated by a combined set of observers that are robust to operating point changes. The outputs of the pre-processing module are the latent residuals which are progressively fed into the decision module, a dynamic weighting ensemble of fault classifiers that incrementally learns the residuals-faults relationships and dynamically classifies the faults including multiple new classes. The pre-processing module consists of the Wold cross-validation algorithm along with the non-linear iterative partial least squares (NIPALS) that projects the residual to the new feature space, extracts the latent information among the residuals and estimates the optimal number of principal components to form the latent residuals. Simulation results confirm the effectiveness of this approach, even in the incomplete scenarios, i.e., the missing data in the batches of generated residuals due to sensor failures.

NEURO-FUZZY BASED FAULT DIAGNOSIS OF A STEAM GENERATOR

Abstract This paper focuses on the development and application of a Neuro-Fuzzy (NF) networks-based scheme for Fault Detection and Isolation (FDI) in a U-tube Steam Generator (UTSG). First, a NF network is trained with data collected from a full scale UTSG simulator, and residuals are generated for fault detection. To identify the UTSG, a Locally Linear Neuro-Fuzzy (LLNF) model is used. This model is trained using the Locally Linear Model Tree (LOLIMOT) algorithm which is an incremental tree structure algorithm. Then, an evolutionary algorithm is used to train a Mamdani type NF network to classify the residuals. The residuals are analyzed by using this NF classifier for fault isolation purposes.

Optimal detection of new classes of faults by an Invasive Weed Optimization method

Proper detection of unknown patterns plays an important role in diagnosing new classes of faults. This can be done by incremental learning of novel information and updating the diagnostic system by appending newly trained fault classifiers in an ensemble design. We consider a new-class fault detector previously developed by the authors and based on thresholding the normalized weighted average of the outputs (NWAO) of the base classifiers in a multi-classifier diagnostic system. A proper tuning of the thresholds in the NWAO detector is necessary to achieve a satisfactory performance. This is done in this paper by specifically introducing a performance function and optimizing it within the necessary trade-off between new class false alarm and new class missed alarm rates, by means of an Invasive Weed Optimization (IWO) algorithm. The optimal NWAO detector is tested with respect to a set of simulated sensor faults in the doubly-fed induction generator (DFIG) of a wind turbine.

Incremental design of a decision system for residual evaluation: A wind turbine application

Abstract This paper presents an incremental way to design the decision module of a diagnostic system by resorting to dynamic weighting ensembles of classifiers. The method is applied for sensor fault detection and isolation in a doubly fed induction generator for a wind turbine application. A bank of observers generates a set of residuals. These signals are progressively fed into a dynamic weighting ensembles algorithm, called Learn++NC, for fault classification. The proposed algorithm incrementally learns the residuals-faults relationships and classifies the faults including multiple new classes, based on a dynamically weighted consult and vote mechanism that combines the outputs of the base-classifiers of the ensemble.

Efficient feature extraction of vibration signals for diagnosing bearing defects in induction motors

This paper presents a model to extract and select a proper set of features for diagnosing bearing defects in induction motors. An efficient pre-processing of the vibration signals is of paramount importance to provide informative features for the fault classification module. The vibration signals are firstly analyzed by the wavelet packet transform to extract informative frequency domain features. The dimension of the set of extracted features is reduced by resorting to linear discriminant analysis to provide a small-size set of informative features for decision making. The fault classification module contains different classifiers that can learn the features-faults relations and classify multiple bearing defects including ball, inner race and outer race defects of different diameters. Experimental results verify the effectiveness of the proposed technique for diagnosing multiple bearing defects in induction motors.

zSlices-Based General Type-2 Fuzzy Fusion of Support Vector Machines With Application to Bearing Fault Detection

This paper proposes a fusion model to enhance classification accuracy of support vector machines (SVMs) for fault detection. The proposed method consists of two different phases, where in the first phase, different SVMs are constructed based on training datasets, and these trained SVMs are evaluated with respect to test datasets by calculating distances between test samples and trained hyperplanes. In order to achieve better results, an optimization scheme based on particle swarm optimization (PSO) is employed to adjust the SVMs parameters. In the next phase, a fusion model, in which the attained accuracies and distances are considered as inputs, is constructed. The fusion model utilizes zSlices-based representation of general type-2 fuzzy logic systems to combine different SVMs. The proposed approach is then applied for bearing fault detection of an induction motor with inner and outer race defects. To investigate the effectiveness of the proposed method, the general type-2 and type-1 fuzzy sets are compared with other two state-of-the-art techniques. The obtained results confirm the superiority of the proposed approach.

Invasive weed classification

Invasive weed optimization (IWO) is a recently published heuristic optimization technique that resembles other evolutionary optimization methods. This paper proposes a new classification technique based on the IWO algorithm, called the invasive weed classification (IWC), to face the problem of pattern classification for multi-class datasets. The aim of the IWC is to find the set of the positions of the class centers that minimize the multi-objective function, i.e., the optimal positions of the class centers. The classification performance is computed as the percentage of misclassified patterns in the testing dataset achieved by the best plants in terms of fitness performance. The performance of the IWC algorithm, both in terms of classification accuracy and training time, is compared with other commonly used classification algorithms.