| Autor: |
Vegad, Swapnil, Panchal, Jaimin, Bhavsar, Keval, Parmar, Umang |
| Předmět: |
|
| Zdroj: |
AIP Conference Proceedings; 2024, Vol. 2960 Issue 1, p1-11, 11p |
| Abstrakt: |
In the last few years, analysis of vibration signals has shown promising results in the early identification of faults in the ball bearings. One of the major challenges is to remove the random background noise using the signal processing technique from the captured vibration signal so that fault prediction can be done with fewer miss-classifications. This study uses the publicly available dataset consisting vibration signal of the bearing data centre at Case Western University. To remove the random background noise from the vibration signal, signal processing techniques namely, Walsh Hadamard Transform (WHT) and Maximum Overlap Discrete Wavelet Transform (MODWT) are used. For denoising the signal using MODWT, a base wavelet is required, which is selected using the criterion namely Maximum Energy to Shannon Entropy (MESE). Using the extracted coefficients of both the signal processing techniques various statistical features are extracted, which are then fed to ReliefF (RF) and Mutual Information (MI) for the selection of important features. A feature vector is developed using the selected features that are then fed to various machine learning models for the identification of various faults. The result shows that the Support Vector Machine with the features selected using Mutual Information accurately identifies the majority of the faults. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
