Induction motor bearing fault diagnostics using i-kaz™ and decision tree classification

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

In this paper, the I-kaz™ method is proposed for the detection of induction motor bearing faults using vibration signals, which has not been presented so far. The purpose of the study is to compare this new technique with the classical kurtosis method in the time domain and to validate the performance of the proposed I-kaz coefficient using a decision tree classification. Three bearing conditions are investigated; i.e. normal, ball fault and inner race fault; with a very small fault size (0.1778 mm). All faulty bearings are artificially damaged using electro-discharge machining and placed in the motor drive end side. The experimental test rig consists of a 2 HP induction motor, a torque transducer, a dynamometer and control electronics. Vibration data is obtained using an accelerometer and analyzed using MATLAB software for the time domain analysis which include the I-kaz graphical and coefficient comparison with time waveform and kurtosis values for all bearing conditions. Then, both features are used to train a conditional inference tree (CTree) fault classifier separately. The proposed I-kaz coefficient provides higher percentage differences between all faulty and normal bearings compared to kurtosis. However, the I-kaz graphic presents similar identification as the time waveform where only the inner race fault is distinguished from the normal bearing. The training classification results also revealed that the I-kaz coefficient is significantly better with an accuracy of 99.64% and a Kappa value of 0.9946 compared to kurtosis at only 63.57% and 0.4536, respectively. Furthermore, all test data were classified accordingly using the I-kaz coefficient whereas for kurtosis, only 65% are correctly classified with the 0.475 Kappa value. It is proved that the I-kaz™ method is suitable for induction bearing fault detection and is recommended as a classification feature, especially for the diagnostics of ball fault which is the most difficult to diagnose.

Original languageEnglish
Pages (from-to)3361-3372
Number of pages12
JournalInternational Journal of Automotive and Mechanical Engineering
Volume13
Issue number2
DOIs
Publication statusPublished - 2016

Fingerprint

Bearings (structural)
Decision trees
Induction motors
Time domain analysis
Electric discharge machining
Dynamometers
Accelerometers
Fault detection
MATLAB
Transducers
Classifiers
Electronic equipment
Torque

Keywords

  • Bearing fault diagnosis
  • Condition monitoring
  • Decision tree classification
  • I-kaz™
  • Vibration signal

ASJC Scopus subject areas

  • Automotive Engineering
  • Mechanical Engineering

Cite this

@article{f3413887cc404397a928218b20897ebc,
title = "Induction motor bearing fault diagnostics using i-kaz™ and decision tree classification",
abstract = "In this paper, the I-kaz™ method is proposed for the detection of induction motor bearing faults using vibration signals, which has not been presented so far. The purpose of the study is to compare this new technique with the classical kurtosis method in the time domain and to validate the performance of the proposed I-kaz coefficient using a decision tree classification. Three bearing conditions are investigated; i.e. normal, ball fault and inner race fault; with a very small fault size (0.1778 mm). All faulty bearings are artificially damaged using electro-discharge machining and placed in the motor drive end side. The experimental test rig consists of a 2 HP induction motor, a torque transducer, a dynamometer and control electronics. Vibration data is obtained using an accelerometer and analyzed using MATLAB software for the time domain analysis which include the I-kaz graphical and coefficient comparison with time waveform and kurtosis values for all bearing conditions. Then, both features are used to train a conditional inference tree (CTree) fault classifier separately. The proposed I-kaz coefficient provides higher percentage differences between all faulty and normal bearings compared to kurtosis. However, the I-kaz graphic presents similar identification as the time waveform where only the inner race fault is distinguished from the normal bearing. The training classification results also revealed that the I-kaz coefficient is significantly better with an accuracy of 99.64{\%} and a Kappa value of 0.9946 compared to kurtosis at only 63.57{\%} and 0.4536, respectively. Furthermore, all test data were classified accordingly using the I-kaz coefficient whereas for kurtosis, only 65{\%} are correctly classified with the 0.475 Kappa value. It is proved that the I-kaz™ method is suitable for induction bearing fault detection and is recommended as a classification feature, especially for the diagnostics of ball fault which is the most difficult to diagnose.",
keywords = "Bearing fault diagnosis, Condition monitoring, Decision tree classification, I-kaz™, Vibration signal",
author = "Othman, {M. S.} and Nuawi, {Mohd. Zaki} and Ramizi Mohamed",
year = "2016",
doi = "10.15282/ijame.13.2.2016.6.0278",
language = "English",
volume = "13",
pages = "3361--3372",
journal = "International Journal of Automotive and Mechanical Engineering",
issn = "2229-8649",
publisher = "The Automotive Engineering Centre (AEC), Universiti Malaysia Pahang",
number = "2",

}

TY - JOUR

T1 - Induction motor bearing fault diagnostics using i-kaz™ and decision tree classification

AU - Othman, M. S.

AU - Nuawi, Mohd. Zaki

AU - Mohamed, Ramizi

PY - 2016

Y1 - 2016

N2 - In this paper, the I-kaz™ method is proposed for the detection of induction motor bearing faults using vibration signals, which has not been presented so far. The purpose of the study is to compare this new technique with the classical kurtosis method in the time domain and to validate the performance of the proposed I-kaz coefficient using a decision tree classification. Three bearing conditions are investigated; i.e. normal, ball fault and inner race fault; with a very small fault size (0.1778 mm). All faulty bearings are artificially damaged using electro-discharge machining and placed in the motor drive end side. The experimental test rig consists of a 2 HP induction motor, a torque transducer, a dynamometer and control electronics. Vibration data is obtained using an accelerometer and analyzed using MATLAB software for the time domain analysis which include the I-kaz graphical and coefficient comparison with time waveform and kurtosis values for all bearing conditions. Then, both features are used to train a conditional inference tree (CTree) fault classifier separately. The proposed I-kaz coefficient provides higher percentage differences between all faulty and normal bearings compared to kurtosis. However, the I-kaz graphic presents similar identification as the time waveform where only the inner race fault is distinguished from the normal bearing. The training classification results also revealed that the I-kaz coefficient is significantly better with an accuracy of 99.64% and a Kappa value of 0.9946 compared to kurtosis at only 63.57% and 0.4536, respectively. Furthermore, all test data were classified accordingly using the I-kaz coefficient whereas for kurtosis, only 65% are correctly classified with the 0.475 Kappa value. It is proved that the I-kaz™ method is suitable for induction bearing fault detection and is recommended as a classification feature, especially for the diagnostics of ball fault which is the most difficult to diagnose.

AB - In this paper, the I-kaz™ method is proposed for the detection of induction motor bearing faults using vibration signals, which has not been presented so far. The purpose of the study is to compare this new technique with the classical kurtosis method in the time domain and to validate the performance of the proposed I-kaz coefficient using a decision tree classification. Three bearing conditions are investigated; i.e. normal, ball fault and inner race fault; with a very small fault size (0.1778 mm). All faulty bearings are artificially damaged using electro-discharge machining and placed in the motor drive end side. The experimental test rig consists of a 2 HP induction motor, a torque transducer, a dynamometer and control electronics. Vibration data is obtained using an accelerometer and analyzed using MATLAB software for the time domain analysis which include the I-kaz graphical and coefficient comparison with time waveform and kurtosis values for all bearing conditions. Then, both features are used to train a conditional inference tree (CTree) fault classifier separately. The proposed I-kaz coefficient provides higher percentage differences between all faulty and normal bearings compared to kurtosis. However, the I-kaz graphic presents similar identification as the time waveform where only the inner race fault is distinguished from the normal bearing. The training classification results also revealed that the I-kaz coefficient is significantly better with an accuracy of 99.64% and a Kappa value of 0.9946 compared to kurtosis at only 63.57% and 0.4536, respectively. Furthermore, all test data were classified accordingly using the I-kaz coefficient whereas for kurtosis, only 65% are correctly classified with the 0.475 Kappa value. It is proved that the I-kaz™ method is suitable for induction bearing fault detection and is recommended as a classification feature, especially for the diagnostics of ball fault which is the most difficult to diagnose.

KW - Bearing fault diagnosis

KW - Condition monitoring

KW - Decision tree classification

KW - I-kaz™

KW - Vibration signal

UR - http://www.scopus.com/inward/record.url?scp=85008238127&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85008238127&partnerID=8YFLogxK

U2 - 10.15282/ijame.13.2.2016.6.0278

DO - 10.15282/ijame.13.2.2016.6.0278

M3 - Article

AN - SCOPUS:85008238127

VL - 13

SP - 3361

EP - 3372

JO - International Journal of Automotive and Mechanical Engineering

JF - International Journal of Automotive and Mechanical Engineering

SN - 2229-8649

IS - 2

ER -