An improved wrapper-based feature selection method for machinery fault diagnosis.
Kar Hoou Hui1, Ching Sheng Ooi1, Meng Hee Lim1
1Institute of Noise and Vibration, Universiti Teknologi Malaysia, Kuala Lumpur, Malaysia.
Plos One
|December 21, 2017
Summary
This study introduces an improved wrapper-based feature selection (WFS) method for machinery fault diagnosis. The new WFS method efficiently identifies the best features for machine learning models, reducing computational effort in diagnosing bearing faults.
Area of Science:
- Mechanical Engineering
- Data Science
- Machine Learning
Background:
- Machinery fault diagnosis traditionally relies heavily on expert interpretation of vibration signals.
- Machine learning offers an alternative but requires effective feature selection for optimal performance.
- Wrapper-based feature selection (WFS) methods face a trade-off between accuracy and computational cost.
Purpose of the Study:
- To propose an improved WFS technique for machinery fault diagnosis.
- To integrate the proposed WFS with a support vector machine (SVM) classifier.
- To evaluate the efficiency and capability of the WFS method using a rolling element bearing dataset.
Main Methods:
- Developed an improved wrapper-based feature selection (WFS) technique.
- Integrated the WFS method with a support vector machine (SVM) model for fault classification.
- Utilized the Case Western Reserve University Bearing Data Centre's vibration dataset for validation.
Main Results:
- The proposed WFS method identified the optimal feature subset with reduced computational effort.
- The WFS technique effectively eliminated redundant feature re-evaluation.
- The integrated WFS-SVM system demonstrated high capability and efficiency in fault diagnosis.
Conclusions:
- The improved WFS technique is capable and efficient for machinery fault diagnosis.
- This approach enhances machine learning model performance by optimizing feature selection.
- The study provides a robust system for diagnosing rolling element bearing faults.
