The induction motor is a key component to provide horsepower and speed to many kinds of industrial machines. Bearing, as one of the most important components in an induction motor, accounts for more than 40% of the total failure modes of induction motors. Thus, to extend the service life of the induction motor and avoid catastrophic economic loss, condition monitoring of induction motor bearings is of great importance in daily maintenance. The vibration-based method is one of the most welcomed methods for monitoring the state of motor bearing failures. However, vibration signal always contains tremendous background noise, such as other rotating frequency components and motor electromagnetic noise, which makes condition monitoring quite challenging work. Besides, the bearing characteristic frequencies which is one of the key parameters for diagnosing bearing faults are always subject to the effect of spectral smearing due to speed variation and nonlinear resonant distortion. This leads to a relatively worse signal- to-noise ratio of the fault signals. To accurately monitor the condition of a motor bearing, an effective technique is needed to de-noise the background noise and eliminate the effect of spectral smearing and resonant distortion. Thus, this study aims to develop an advanced technique based on multi-scale analysis of residual vibration signals of induction motors to achieve diagnostic of bearing failures. In this study, firstly, the frequency change caused by spectral smearing and resonant distortion are studied to gain insightful understanding of the generation and behaviour with operating conditions. Then, the residual signal of the motor vibration is achieved by removing the high-amplitude frequency components from the spectrum. This is achieved by the Gini Index (GI) which finds the optimal threshold in a series of thresholds. Moreover, the multi-scale analysis based on Wavelet Decomposition (WD) was applied to the residual signal for further noise suppression. Lastly, the envelope analysis was performed on all 4 types of signals: raw vibration signal, residual signal, the signal after applying WD to the raw signal, and the signal after applying WD to the residual signal, to get the characteristic frequencies for fault diagnosis. To validate the performances of these types of signals, a test rig was set up to conduct a number of tests with different induction motor bearing faults. The experimental study demonstrates that comparing all 4 types of signals, the proposed combined method of residual signal and Wavelet Decomposition can successfully suppress the noise and keep the frequency components relevant to the fault, which significantly increase the SNR for a trustworthy fault diagnosis.
| Date of Award | 4 Oct 2024 |
|---|
| Original language | English |
|---|
| Supervisor | Fengshou Gu (Main Supervisor) & Andrew Ball (Co-Supervisor) |
|---|