Development of novel intelligent condition monitoring procedures for rolling element bearings
The primary aim of this thesis is to develop a novel procedure for an intelligent automatic diagnostic condition monitoring system for rolling element bearings. The applicability of this procedure is demonstrated by its implementation in a particular electric motor drive system. The novel bearing condition diagnostic procedure developed involves three stages combining the merits of advanced signal processing techniques, feature extraction methods and artificial neural networks. This procedure is the effective combination of these techniques and methods in a holistic approach to the rolling element bearing problem which provides the novelty in this thesis. Maintenance costs account for an extremely large proportion of the operating costs of machinery. In addition, machine breakdowns and consequent downtime can severely affect the productivity of factories and the safety of products. It is therefore becoming increasingly important for industries to monitor their equipment systematically in order to reduce the number of breakdowns and to avoid unnecessary costs and delays caused by repair. The rolling element bearing is an extremely widespread component in industrial rotating machinery and a large number of problems arise from faulty bearings. Therefore, proper monitoring of bearing condition is highly cost-effective in reducing operating cost. The advanced signal processing techniques used here are bispectral-based and wavelet-based analyses. The bispectral-based procedures examined are the bis-pectrum, the bicoherence, the bispectrum diagonal slice, the bicoherence diagonal slice, the summed bispectrum and the summed bicoherence. The wavelet-based procedure uses the Morlet wavelet. These methods greatly enhance the ability of an automated diagnostic process by linking the increased capability for signal analysis to the predictive capability of artificial neural networks. The bearing monitoring scheme based on bispectral analysis is shown to provide greater insight into the structure of bearing vibration signals and to offer more diagnostic information than conventional power spectral analysis. The wavelet analysis provides a multi-resolution, time-frequency approach to extract information from the bearing vibration signatures. In order to effectively interpret the wavelet map, the time-frequency domain is used instead of the time-scale domain by plotting the associated time trace and power spectrum.