Signal Processing for Fault Detection and Diagnosis in Electric Machines and Systems
Over the last three decades, the search for competitiveness and growth gains has driven the evolution of machine maintenance policies, and the industry has moved from passive maintenance to active maintenance with the aim of improving productivity. Active maintenance requires continuous monitoring of industrial systems in order to increase reliability, availability rates and guarantee the safety of people and property. This book presents the main advanced signal processing techniques for fault detection and diagnosis in electromechanical systems. It focuses on presenting these advanced tools from time-frequency representation and time-scale analysis to demodulation techniques, including innovative and recently developed options. Each technique is evaluated and compared, and its advantages and drawbacks highlighted. Parametric spectral analysis, which aims to handle some of the main drawbacks of these approaches, is introduced as a potential solution. Signal Processing for Fault Detection and Diagnosis in Electric Machines and Systems offers thorough, analytical coverage of the following topics: parametric signal processing approach; the signal demodulation techniques; Kullback-Leibler divergence for incipient fault diagnosis; high-order spectra (HOS); and fault detection and diagnosis based on principal component analysis. Finally, a brief conclusion suggests some possibilities for the future direction of the field. The book is a useful resource for researchers and engineers whose work involves electrical machines or fault detection specifically, and also of value to postgraduate students with an interest in entering this field.
Inspec keywords: principal component analysis; electric machines; demodulation; fault diagnosis; signal processing
Other keywords: fault detection; higher-order spectra; Kullback-Leibler divergence; signal demodulation techniques; parametric signal processing approach; principal component analysis; electric machines; incipient fault diagnosis
Subjects: General and management topics; Other topics in statistics; General electrical engineering topics; a.c. machines; Signal processing and detection; Digital signal processing; d.c. machines; Other topics in statistics
- Book DOI: 10.1049/PBPO153E
- Chapter DOI: 10.1049/PBPO153E
- ISBN: 9781785619571
- e-ISBN: 9781785619588
- Page count: 284
- Format: PDF
-
Front Matter
- + Show details - Hide details
-
p.
(1)
-
Introduction
- + Show details - Hide details
-
p.
1
–2
(2)
-
1 Parametric signal processing approach
- + Show details - Hide details
-
p.
3
–50
(48)
This chapter addresses the issue of condition monitoring based on MCSA using parametric spectral estimation techniques and detection theory. This approach is used for fault characteristics estimation. Then, generalized likelihood ratio test (GLRT) is used for automatic decision-making. The proposed fault detection approach uses fault frequency signature bins and amplitude estimators, and a fault decision module based on statistical tools. MLE is used for fault characteristics computation. Then, composite hypothesis testing is used as a decision module. The main objective is to discriminate the healthy induction motor from a faulty one. Finally, a fault severity measurement criterion is proposed and demonstrated for several induction motor fault detection.
-
2 The signal demodulation techniques
- + Show details - Hide details
-
p.
51
–83
(33)
Condition monitoring of electrical machines is a broad scientific area, the ultimate purpose of which is to ensure the safe, reliable and continuous operation of electrical machines. Hence, the task of fault detection is still an art, because induction machines are widely used in variable speed drives and in renewable energy conversion systems. A deep knowledge about all the phenomena involved during the occurrence of a failure constitutes an essential background for the development of any failure detection and diagnosis system. For the failure detection problem, it is important to know if a failure exists or not in the electric machine via the processing of available measurements. This chapter provides then an approach based on a electric machine current data collection and attempts to highlight the use of demodulation techniques for failure detection for stationary and nonstationary cases.
-
3 Kullback–Leibler divergence for incipient fault diagnosis
- + Show details - Hide details
-
p.
87
–118
(32)
This chapter discusses the issue of incipient fault detection and diagnosis (FDD). After a general introduction, the requirements for FDD methods are defined under the three criteria of robustness, sensitivity, and simplicity. A methodology of FDD is also introduced in four main steps: modelling, preprocessing, features extraction, and features analysis. After the definition of incipient fault based on the levels of fault, signal, and environmental nuisances, a paradigm is drawn between information-hiding domain and FDD. We will show that dissimilarity measure of probability density function (PDF) used for data hiding is efficient for incipient fault detection. The methodology is illustrated through incipient crack detection in a conductive material using eddy currents and short intermittent open-circuit duration in three-level neutralpoint-clamped inverter. The chapter also discusses fault detection threshold optimal setting and fault severity estimation.
-
4 Higher-order spectra
- + Show details - Hide details
-
p.
119
–202
(84)
Over the past decades, higher-order spectra (HOS), also called polyspectra, have established a status as a suitable mathematical and signal processing tool for nonlinear system analysis. However, a major problem with this kind of signal processing tool application is the interpretation of the obtained results, and much uncertainty still exists about the relation between HOS contribution compared with the second-order statistics. This chapter provided an important opportunity to advance the understanding advantages of HOS. The classical power spectrum (PS), which is defined as the Fourier transform (FT) of the autocorrelation sequence (the second-order cumulant), does not give any information about the phase of system frequency response; therefore, it is unable to give any indication about system nonlinearity. However, the HOS [1-11] are defined as the multidimensional FT of higher-order cumulants of a stationary random process and can overcome the inability of PS to detect these nonlinearities.
-
5 Fault detection and diagnosis based on principal component analysis
- + Show details - Hide details
-
p.
203
–258
(56)
In this chapter, PCA, relative PCA (RPCA) [37,38] and normalization PCA (NPCA) are introduced with application in fault detection and fault diagnosis. There are some theories and applications about PCA, such as the basic principles of PCA, geometrical interpretation of PCA, Hotelling's T2 statistic and SPE statistic for fault detection's control limit. Then a fault detection method based on PCA is introduced for Tennessee Eastman (TE) process. What's more, the fault diagnosis method based on PCA is introduced with its application for inverter. There are some theories and application about RPCA, such as the definition of Relative Transform, basic principles of RPCA and geometrical interpretation of RPCA. Then the fault detection method based on RPCA is introduced with its application. In addition, in order to improve the control limit of PCA with Hotelling's T2, the dynamic data window control limit algorithm based on RPCA is introduced with its application. As follows, the fault diagnosis method based on RPCA is introduced with its application. There are some theories and application about NPCA, such as the definition of longitudinal standardization (LS) and basic principles of NPCA. Next a fault detection method based on NPCA is presented with its application in wind power generation. Then another fault detection method based on NPCA is presented with its application in DC motor. In order to increase the control limit of PCA with Hotelling's T2, a fault detection method based on NPCA-adaptive confidence limit (ACL) is presented with application in DC motor.
-
Conclusion
- + Show details - Hide details
-
p.
259
–260
(2)
In summary, this book has identified opportunities of some advanced signal processing techniques for electromechanical systems fault detection and diagnosis. It has provided methodologies and algorithms with several illustrative examples and practical case studies, while highlighting some prospective investigations.
-
Back Matter
- + Show details - Hide details
-
p.
(1)