Application of power spectrum, cepstrum, higher order spectrum and neural network analyses for induction motor fault diagnosis

Research output: Contribution to journalArticle

58 Citations (Scopus)

Abstract

The power spectrum is defined as the square of the magnitude of the Fourier transform (FT) of a signal. The advantage of FT analysis is that it allows the decomposition of a signal into individual periodic frequency components and establishes the relative intensity of each component. It is the most commonly used signal processing technique today. If the same principle is applied for the detection of periodicity components in a Fourier spectrum, the process is called the cepstrum analysis. Cepstrum analysis is a very useful tool for detection families of harmonics with uniform spacing or the families of sidebands commonly found in gearbox, bearing and engine vibration fault spectra. Higher order spectra (HOS) (also known as polyspectra) consist of higher order moment of spectra which are able to detect non-linear interactions between frequency components. For HOS, the most commonly used is the bispectrum. The bispectrum is the third-order frequency domain measure, which contains information that standard power spectral analysis techniques cannot provide. It is well known that neural networks can represent complex non-linear relationships, and therefore they are extremely useful for fault identification and classification. This paper presents an application of power spectrum, cepstrum, bispectrum and neural network for fault pattern extraction of induction motors. The potential for using the power spectrum, cepstrum, bispectrum and neural network as a means for differentiating between healthy and faulty induction motor operation is examined. A series of experiments is done and the advantages and disadvantages between them are discussed. It has been found that a combination of power spectrum, cepstrum and bispectrum plus neural network analyses could be a very useful tool for condition monitoring and fault diagnosis of induction motors.
Original languageEnglish
Pages (from-to)342-360
Number of pages19
JournalMechanical Systems and Signal Processing
Volume39
Issue number1-2
DOIs
Publication statusPublished - Aug 2013

Fingerprint

Power spectrum
Induction motors
Failure analysis
Neural networks
Fourier transforms
Bearings (structural)
Condition monitoring
Spectrum analysis
Signal processing
Engines
Decomposition
Experiments

Cite this

@article{3cb0543901e947c8afcb3b8ce432eb85,
title = "Application of power spectrum, cepstrum, higher order spectrum and neural network analyses for induction motor fault diagnosis",
abstract = "The power spectrum is defined as the square of the magnitude of the Fourier transform (FT) of a signal. The advantage of FT analysis is that it allows the decomposition of a signal into individual periodic frequency components and establishes the relative intensity of each component. It is the most commonly used signal processing technique today. If the same principle is applied for the detection of periodicity components in a Fourier spectrum, the process is called the cepstrum analysis. Cepstrum analysis is a very useful tool for detection families of harmonics with uniform spacing or the families of sidebands commonly found in gearbox, bearing and engine vibration fault spectra. Higher order spectra (HOS) (also known as polyspectra) consist of higher order moment of spectra which are able to detect non-linear interactions between frequency components. For HOS, the most commonly used is the bispectrum. The bispectrum is the third-order frequency domain measure, which contains information that standard power spectral analysis techniques cannot provide. It is well known that neural networks can represent complex non-linear relationships, and therefore they are extremely useful for fault identification and classification. This paper presents an application of power spectrum, cepstrum, bispectrum and neural network for fault pattern extraction of induction motors. The potential for using the power spectrum, cepstrum, bispectrum and neural network as a means for differentiating between healthy and faulty induction motor operation is examined. A series of experiments is done and the advantages and disadvantages between them are discussed. It has been found that a combination of power spectrum, cepstrum and bispectrum plus neural network analyses could be a very useful tool for condition monitoring and fault diagnosis of induction motors.",
keywords = "Bispectrum, Cepstrum, Fault pattern identification, Induction motors, Power spectrum, Vibration",
author = "B. Liang and Iwnicki, {S. D.} and Y. Zhao",
year = "2013",
month = "8",
doi = "10.1016/j.ymssp.2013.02.016",
language = "English",
volume = "39",
pages = "342--360",
journal = "Mechanical Systems and Signal Processing",
issn = "0888-3270",
publisher = "Academic Press Inc.",
number = "1-2",

}

TY - JOUR

T1 - Application of power spectrum, cepstrum, higher order spectrum and neural network analyses for induction motor fault diagnosis

AU - Liang, B.

AU - Iwnicki, S. D.

AU - Zhao, Y.

PY - 2013/8

Y1 - 2013/8

N2 - The power spectrum is defined as the square of the magnitude of the Fourier transform (FT) of a signal. The advantage of FT analysis is that it allows the decomposition of a signal into individual periodic frequency components and establishes the relative intensity of each component. It is the most commonly used signal processing technique today. If the same principle is applied for the detection of periodicity components in a Fourier spectrum, the process is called the cepstrum analysis. Cepstrum analysis is a very useful tool for detection families of harmonics with uniform spacing or the families of sidebands commonly found in gearbox, bearing and engine vibration fault spectra. Higher order spectra (HOS) (also known as polyspectra) consist of higher order moment of spectra which are able to detect non-linear interactions between frequency components. For HOS, the most commonly used is the bispectrum. The bispectrum is the third-order frequency domain measure, which contains information that standard power spectral analysis techniques cannot provide. It is well known that neural networks can represent complex non-linear relationships, and therefore they are extremely useful for fault identification and classification. This paper presents an application of power spectrum, cepstrum, bispectrum and neural network for fault pattern extraction of induction motors. The potential for using the power spectrum, cepstrum, bispectrum and neural network as a means for differentiating between healthy and faulty induction motor operation is examined. A series of experiments is done and the advantages and disadvantages between them are discussed. It has been found that a combination of power spectrum, cepstrum and bispectrum plus neural network analyses could be a very useful tool for condition monitoring and fault diagnosis of induction motors.

AB - The power spectrum is defined as the square of the magnitude of the Fourier transform (FT) of a signal. The advantage of FT analysis is that it allows the decomposition of a signal into individual periodic frequency components and establishes the relative intensity of each component. It is the most commonly used signal processing technique today. If the same principle is applied for the detection of periodicity components in a Fourier spectrum, the process is called the cepstrum analysis. Cepstrum analysis is a very useful tool for detection families of harmonics with uniform spacing or the families of sidebands commonly found in gearbox, bearing and engine vibration fault spectra. Higher order spectra (HOS) (also known as polyspectra) consist of higher order moment of spectra which are able to detect non-linear interactions between frequency components. For HOS, the most commonly used is the bispectrum. The bispectrum is the third-order frequency domain measure, which contains information that standard power spectral analysis techniques cannot provide. It is well known that neural networks can represent complex non-linear relationships, and therefore they are extremely useful for fault identification and classification. This paper presents an application of power spectrum, cepstrum, bispectrum and neural network for fault pattern extraction of induction motors. The potential for using the power spectrum, cepstrum, bispectrum and neural network as a means for differentiating between healthy and faulty induction motor operation is examined. A series of experiments is done and the advantages and disadvantages between them are discussed. It has been found that a combination of power spectrum, cepstrum and bispectrum plus neural network analyses could be a very useful tool for condition monitoring and fault diagnosis of induction motors.

KW - Bispectrum

KW - Cepstrum

KW - Fault pattern identification

KW - Induction motors

KW - Power spectrum

KW - Vibration

UR - http://www.scopus.com/inward/record.url?scp=84879692738&partnerID=8YFLogxK

U2 - 10.1016/j.ymssp.2013.02.016

DO - 10.1016/j.ymssp.2013.02.016

M3 - Article

VL - 39

SP - 342

EP - 360

JO - Mechanical Systems and Signal Processing

JF - Mechanical Systems and Signal Processing

SN - 0888-3270

IS - 1-2

ER -