Pattern recognition of mechanical fault signal using hidden markov model

Jong Min Lee, Seung-Jong Kim, Yoha Hwang, Chang Seop Song

Research output: Contribution to conferencePaper

1 Citation (Scopus)

Abstract

Hidden Markov Model (HMM) is a very powerful signal analysis method. However, despite its huge success in voice recognition, its use in mechanical engineering has been very limited. In this paper, HMM algorithm has been optimized to be used in mechanical signal analysis and was successfully applied to recognition of rotor fault patterns. Both continuous (CHMM) and discrete (DHMM) types were studied. Rotor kit was set under unbalance and oil whirl conditions and time signals of two failure conditions were sampled and converted to auto-power spectrums. Using filter bank, feature vectors were extracted from auto-power spectrum. After CHMMs and DHMMs were trained, each HMM was applied to all sampled data and has shown accurate rotor fault recognition ability. HMM has shown good recognition ability in spite of small number of training data set. Judging by the result, CHMM has better recognition ability and DHMM has more robust recognition ability in this study.

Original languageEnglish
Pages4723-4730
Number of pages8
Publication statusPublished - 2003 Dec 1
Externally publishedYes
EventProceedings of the Tenth International Congress on Sound and Vibration - Stockholm, Sweden
Duration: 2003 Jul 72003 Jul 10

Conference

ConferenceProceedings of the Tenth International Congress on Sound and Vibration
CountrySweden
CityStockholm
Period03/7/703/7/10

Fingerprint

Hidden Markov models
Pattern recognition
Rotors
Signal analysis
Power spectrum
Filter banks
Mechanical engineering
Speech recognition

ASJC Scopus subject areas

  • Engineering(all)

Cite this

Lee, J. M., Kim, S-J., Hwang, Y., & Song, C. S. (2003). Pattern recognition of mechanical fault signal using hidden markov model. 4723-4730. Paper presented at Proceedings of the Tenth International Congress on Sound and Vibration, Stockholm, Sweden.

Pattern recognition of mechanical fault signal using hidden markov model. / Lee, Jong Min; Kim, Seung-Jong; Hwang, Yoha; Song, Chang Seop.

2003. 4723-4730 Paper presented at Proceedings of the Tenth International Congress on Sound and Vibration, Stockholm, Sweden.

Research output: Contribution to conferencePaper

Lee, JM, Kim, S-J, Hwang, Y & Song, CS 2003, 'Pattern recognition of mechanical fault signal using hidden markov model' Paper presented at Proceedings of the Tenth International Congress on Sound and Vibration, Stockholm, Sweden, 03/7/7 - 03/7/10, pp. 4723-4730.
Lee JM, Kim S-J, Hwang Y, Song CS. Pattern recognition of mechanical fault signal using hidden markov model. 2003. Paper presented at Proceedings of the Tenth International Congress on Sound and Vibration, Stockholm, Sweden.
Lee, Jong Min ; Kim, Seung-Jong ; Hwang, Yoha ; Song, Chang Seop. / Pattern recognition of mechanical fault signal using hidden markov model. Paper presented at Proceedings of the Tenth International Congress on Sound and Vibration, Stockholm, Sweden.8 p.
@conference{be21fef183864e55ac22a92b5f6307a0,
title = "Pattern recognition of mechanical fault signal using hidden markov model",
abstract = "Hidden Markov Model (HMM) is a very powerful signal analysis method. However, despite its huge success in voice recognition, its use in mechanical engineering has been very limited. In this paper, HMM algorithm has been optimized to be used in mechanical signal analysis and was successfully applied to recognition of rotor fault patterns. Both continuous (CHMM) and discrete (DHMM) types were studied. Rotor kit was set under unbalance and oil whirl conditions and time signals of two failure conditions were sampled and converted to auto-power spectrums. Using filter bank, feature vectors were extracted from auto-power spectrum. After CHMMs and DHMMs were trained, each HMM was applied to all sampled data and has shown accurate rotor fault recognition ability. HMM has shown good recognition ability in spite of small number of training data set. Judging by the result, CHMM has better recognition ability and DHMM has more robust recognition ability in this study.",
author = "Lee, {Jong Min} and Seung-Jong Kim and Yoha Hwang and Song, {Chang Seop}",
year = "2003",
month = "12",
day = "1",
language = "English",
pages = "4723--4730",
note = "Proceedings of the Tenth International Congress on Sound and Vibration ; Conference date: 07-07-2003 Through 10-07-2003",

}

TY - CONF

T1 - Pattern recognition of mechanical fault signal using hidden markov model

AU - Lee, Jong Min

AU - Kim, Seung-Jong

AU - Hwang, Yoha

AU - Song, Chang Seop

PY - 2003/12/1

Y1 - 2003/12/1

N2 - Hidden Markov Model (HMM) is a very powerful signal analysis method. However, despite its huge success in voice recognition, its use in mechanical engineering has been very limited. In this paper, HMM algorithm has been optimized to be used in mechanical signal analysis and was successfully applied to recognition of rotor fault patterns. Both continuous (CHMM) and discrete (DHMM) types were studied. Rotor kit was set under unbalance and oil whirl conditions and time signals of two failure conditions were sampled and converted to auto-power spectrums. Using filter bank, feature vectors were extracted from auto-power spectrum. After CHMMs and DHMMs were trained, each HMM was applied to all sampled data and has shown accurate rotor fault recognition ability. HMM has shown good recognition ability in spite of small number of training data set. Judging by the result, CHMM has better recognition ability and DHMM has more robust recognition ability in this study.

AB - Hidden Markov Model (HMM) is a very powerful signal analysis method. However, despite its huge success in voice recognition, its use in mechanical engineering has been very limited. In this paper, HMM algorithm has been optimized to be used in mechanical signal analysis and was successfully applied to recognition of rotor fault patterns. Both continuous (CHMM) and discrete (DHMM) types were studied. Rotor kit was set under unbalance and oil whirl conditions and time signals of two failure conditions were sampled and converted to auto-power spectrums. Using filter bank, feature vectors were extracted from auto-power spectrum. After CHMMs and DHMMs were trained, each HMM was applied to all sampled data and has shown accurate rotor fault recognition ability. HMM has shown good recognition ability in spite of small number of training data set. Judging by the result, CHMM has better recognition ability and DHMM has more robust recognition ability in this study.

UR - http://www.scopus.com/inward/record.url?scp=2342430450&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=2342430450&partnerID=8YFLogxK

M3 - Paper

SP - 4723

EP - 4730

ER -