The Application Of Wavelet Analysis To Fault Diagnosis Of Rotating Machinery

A study has been made of applications of wavelet analysis to fault diagnosis of some rotating machinery, including gear, drilling, rolling element bearings and gearbox. A method is presented for local gear fault detection by using discrete dyadic wavelet transform. This scheme avoids the complexities of numerical integration and reduces redundancy of discrete sampling in scale parameter of continuous wavelet transform, hence high in efficiency. A method is proposed for tool wear monitoring in drilling, which is based on the difference of wavelet transform modulus maxima evolution behaviors between the signals that describe the process defects and the process noise. An assessment criterion and a feature extraction procedure are given. Also, this method is tested by practical examples. The discrete dyadic wavelet transform is employed for fault diagnosis of rolling element bearings. Experimental results from tests show that this technique is effective for inner race faults and outer race faults. The singularity with Lipschitz exponents obtained from wavelet transform has been applied to the fault diagnosis of rolling element bearings. We show that there are distinct differences in the singularity measured by Lipschitz exponents with wavelet transforms of the impacts in vibration signals, which are obtained from normal bearings and those with outer, inner and rolling element faults in otherwise identical bearings. This research indicates that singularity measured by Lipschitz exponents can be an index that characterizes the bearing condition and has great potential to be a useful tool for the defect detection in rotating machinery. A new method is proposed for fault detection and identification in gearboxes. The approach is based on wavelet packet transform and fuzzy techniques in pattern recognition. The property of the time-frequency(scale) localization of the wavelet transform is well suited to extracting key features of the short-time variations, caused by a fault such as an outer or inner race spall in bearing vibration signals, thus creation good conditions for pattern recognition. Compared with other time-frequency analysis techniques, wavelet packet transform is of fast algorithm so that this method makes on-line and real-time possible.