Automatic retreading of spectral faulty features from defective rolling element bearings by using reassignment wavelet and higher order statistical analysis

Much work has been done on how to detect fault-development indicators that point to machine breakdowns. Bearing faults are the most frequent causes of failures in rotating machinery. However, signals generated by bearings are often overwhelmed by higher amplitude signals from nearby components or surrounding noise. Useful information is thus lost that the maintenance personnel can hardly perform proper remedies to defective machines. In traditional approaches, Wavelet is one of the commonly adopted tools for analyzing the faulty bearing signal. However, Conventional Wavelet is limited by its energy leakage. To solve this shortcoming, Reassignment Wavelet is introduced in this research to concentrate and localize the defective bearing signal. One of the aims of this research is to use Reassignment method to minimize the energy leakage caused by conventional Wavelet. Also, a Spectrum of Root Mean Square (RMS) and Kurtosis based method is introduced to automatically detect the location of the bearing excitation zone. Finally, through the decomposition analysis, the impact intervals in time can be revealed for determining the kind of bearing faults In this research, a completely new data acquisition system with a well-established virtual instrument has been developed. Comprehensive data was collected from both a laboratory fault simulating machine and industrial machineries. In addition, a new method, Reassignment Wavelet based Spectral RMS x Kurtosis has been developed to facilitate the identification of anomalous impacts appearing in different detective machineries. The results successfully reveal that this method is indicative in detecting the bearing faults at early stage. In particular, with the help of Reassignment Wavelet decomposition and Spectral RMS x Kurtosis, both naturally developed and artificially induced impacts of the defective bearings could be located even though it is presented in a noisy environment. The contribution of this research is the successful development of a more effective signal analysis system for vibration based machine fault diagnosis. By incorporating the new Reassignment Wavelet based Spectral RMS x Kurtosis methodology, the system has been proved to be very useful in helping the industry to uncover machine faults at their early stage of deterioration.