Online incipient fault diagnosis based on Kullback‐Leibler divergence and recursive principle component analysis

Chai, Yi; Tao, Songbing; Mao, Wanbiao; Zhang, Ke; Zhu, Zhiqin

doi:10.1002/cjce.22962

Cited by 20 publications

(6 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Kullback-Leibler Divergence (KLD), or the relative entropy, is a well-known probabilistic tool that has proven its worth in machine learning, neuroscience, pattern recognition [25], and anomaly detection [26,27]. KLD has already proved its efficiency for detecting incipient faults in several applications [28,29].…”

Section: Features Extraction For Fault Classificationmentioning

confidence: 99%

Three-Level NPC Inverter Incipient Fault Detection and Classification using Output Current Statistical Analysis

et al. 2019

View full text Add to dashboard Cite

This paper deals with open switch Fault Detection and Diagnosis (FDD) in three-level Neutral Point Clamped (NPC) inverter for electrical drives. The approach is based on the already available phase current time series measurements for different operating conditions (motor speed, load, and environment noise). Both fault detection and classification are studied and the efficiency performances of the proposed selected features are shown. For the fault detection, we focus on the first four statistical moments and the extracted features and then the Cumulative Sum (CUSUM) algorithm as the feature analysis technique to improve the performances. For the classification study, we propose to couple the knowledge on the faulty system brought by the statistical moments and the Kullback-Leibler divergence particularly suitable for the detection of incipient changes. The Principal Component Analysis (PCA) is then used to perform the classification. A 2D framework is obtained, which allows the faults to be classified efficiently within the considered operating conditions for all the selected fault durations.

show abstract

Section: Features Extraction For Fault Classificationmentioning

confidence: 99%

Three-Level NPC Inverter Incipient Fault Detection and Classification using Output Current Statistical Analysis

et al. 2019

View full text Add to dashboard Cite

show abstract

“…The principal component analysis (PCA), neural network, and support vector machine (SVM) represent the typical data-driven tools for sensor fault diagnoses. PCA methods [7]- [11] can be used to reduce data dimension and extract feature vectors. The PCA-fused algorithms therefore can provide solutions to feature extraction and fault identification for sensors.…”

Section: Introductionmentioning

confidence: 99%

Data Association-Based Fault Diagnosis of IMUs: Optimized DBN Design and Wheeled Robot Evaluation

et al. 2020

View full text Add to dashboard Cite

Deep belief network (DBN) is now being recognized as a powerful and eminently practical tool for large scale data processing. The main characteristics of DBN are the feature extension from low-level content to high-level data association and the representation of joint distribution between original data and matched labels. For a wheeled robot with no other available location reference supports, the internally integrated inertial measurement units (IMUs) essentially requires the robot to be able to implement efficient fault diagnosis to locate and identify the faults, especially for the accumulated error caused by large drifts of gyroscopes. An optimized DBN based fault diagnosis design is proposed to deal with such faults with complexity and diversity. The highlights of the proposed DBN model lies in its combination of weight value optimization via an inexact LSA-GA (abbreviates 'inexact linear searching algorithm-genetic algorithm') and dynamic adjustment for hidden-layer neurons of constituent RBMs (abbreviates 'restricted Boltzmann machines'). The problems associated with DBN anatomy, bat algorithm (BA) description and fault diagnosis modeling are discussed in detail. The real robot platform experiments and dataset tests are conducted. The results indicate that, the optimized DBN design leads to a better fault classification with excellent generalization ability on given datasets, and the adjustable 'DBN structure' contributes to the data association extraction between multiples of fault categories. The proposed scheme may therefore be considered to provide preferred reference models for a class of data based fault diagnosis problems.

show abstract

“…In addition, from the view of probability density function (PDF), the Kullback Leibler divergence-PCA method was implemented by Harmouche et al because of its high sensitivity for incipient faults. Chen et al and Chai et al also discussed the combination of KLD and PCA for incipient fault detection. Considering the extraction of early fault features from process noise, a two-step fault detection method was proposed by Ge et al, which applies wavelet analysis and optimal parity space analysis to distill the characteristics of small amplitude changes.…”

Section: Introductionmentioning

confidence: 99%

Incipient Fault Detection for Chemical Processes Using Two-Dimensional Weighted SLKPCA

Deng

2019

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

Early detection of incipient faults is a challenging task in chemical process monitoring field. As an effective incipient fault detection tool, statistical local kernel principal component analysis (SLKPCA) has demonstrated its advantage over the traditional kernel principal component analysis (KPCA). However, how to improve its incipient fault detection performance is still a valuable problem. In this paper, an enhanced SLKPCA method, referred to as two-dimensional weighted SLKPCA (TWSLKPCA), is proposed by integrating the sample and component weighting strategies. Different to KPCA, SLKPCA monitors the process changes based on the residual vectors computed by the statistical local approach. To highlight the influence of the faulty residual samples, the residual sample weighting strategy is first designed based on the distance between the tested samples and the training samples, which puts large weights on the samples with strong fault information. Furthermore, the residual component weighting strategy is developed to assign large weights to the sensitive components, which are judged by computing the mutual information between the sampleweighted residual components and the original measured variables. Based on the two-dimensional weighted residual vectors, two monitoring statistics are built to detect the incipient faults. Finally, simulations on a numerical example and the Tennessee Eastman process are used to demonstrate the superiority of the proposed TWSLKPCA method.

show abstract

Online incipient fault diagnosis based on Kullback‐Leibler divergence and recursive principle component analysis

Cited by 20 publications

References 28 publications

Three-Level NPC Inverter Incipient Fault Detection and Classification using Output Current Statistical Analysis

Three-Level NPC Inverter Incipient Fault Detection and Classification using Output Current Statistical Analysis

Data Association-Based Fault Diagnosis of IMUs: Optimized DBN Design and Wheeled Robot Evaluation

Incipient Fault Detection for Chemical Processes Using Two-Dimensional Weighted SLKPCA

Contact Info

Product

Resources

About