Abstract

As the most common mechanisms for transmitting power and motion, gears and bearings have been widely used in various mechanical equipment. Many accidents have happened because of failing to detect and replace the faulty gears or bearings in time. Hence it is very important to perform accurate fault diagnosis of gears and bearings. When a gear or bearing has a fault, the vibration signal collected from the mechanical equipment will become non-stationary and contain a series of periodic impulses that are caused by the fault. Many theories and techniques have been developed to extract the faulty information based on analyzing periodic impulses contained in the vibration signals.

It has been reported that neither time-domain analysis nor frequency-domain analysis can do well in analyzing non-stationary signals. Hence time-frequency analysis methods based on wavelet transform and Hilbert-Huang Transform (HHT) have been investigated. Wavelet transform includes continuous wavelet transform (CWT) and discrete wavelet transform (DWT). Research was reported to compare the performance of these methods in fault diagnosis of mechanical components. However, the previous works either only compared HHT based methods with CWT based methods, or only compared HHT based methods with DWT based methods, for certain applications. There are no reported comprehensive comparisons of the three methods for fault diagnosis of gears and bearings.

Cepstrum analysis can detect the periodicity and reduce the influence of the noise for the low energy signals. However, previous research usually only applied Cepstrum analysis directly to extract fault features from the entire original signal. Some other existing methods first applied DWT to decompose the original signal to obtain the detail signals, and then used Hilbert spectral analysis to analyze the periodic impulses contained in the detail signals through identifying faulty characteristic frequency with frequency domain analysis and plotting the instantaneous amplitude with time domain analysis. Hilbert spectral analysis has the advantage in frequency domain analysis. However, sometimes it is not very sensitive in time domain analysis when the energy of the periodic impulses is not strong enough.

In this thesis, we investigate fault diagnosis of gears and bearings using two sets of vibration monitoring data collected in the lab environment: one set for gear condition monitoring and the other for bearing condition monitoring. We propose an improved DWT method that integrates Cepstrum analysis to analyze the periodic impulses contained in the data. With the proposed method, the vibration signals are first decomposed using DWT, and Cepstrum analysis is used to analyze the resulting detail signals. The results show that the proposed method performs better than the existing methods of applying Cepstrum analysis directly. Furthermore, with the help of Cepstrum analysis, the proposed method has better performance in time domain analysis than Hilbert spectral analysis in analyzing the periodic impulses contained in the detail signals.

A comprehensive study is conducted in this thesis to compare the following three methods in fault diagnosis of the gears and bearings: (1) The CWT method using time-wavelet energy spectrum, (2) the improved DWT method using Cepstrum analysis, and (3) the HHT method. The results show that in fault diagnosis of gears, the HHT method has better noise immunity and is more sensitive in frequency domain analysis than the other two methods. The proposed method shows its advantage in analyzing the periodic impulses in time domain than the other two methods. In fault diagnosis of the bearings, the fault can be more clearly detected using the CWT and DWT methods in analyzing the periodic impulse that caused by the outer race fault of the bearing. The results obtained in this thesis can assist researchers and practitioners to select suitable methods for fault diagnosis of gears and bearings.