| The investigation reported in this thesis is concerned with the exploration of Machine Condition Monitoring based on the analyses of transient vibration signals. The Prony method, a modelling technique that fits a linear combination of a series of exponentially decaying sinusoidal functions to sampled data such that the model describes the underlying process of the sampled data, was adopted to construct an analysing procedure as the basis of a useful machine condition monitoring and diagnostic technique.; Issues regarding model order selection were scrutinised. A numerical investigation on the performance of the few well recognised and commonly employed model order selection criteria, the Akaike Information Criteria (AIC), Minimum Descriptive Length (MDL) and Hannan's criterion (Hannan), was carried out. The results of the investigation show that attention should be paid to the order range used for each of the criteria and that care must be taken in the real time application of these criteria. A new model order selection criterion, the normalised error method, was derived. It was demonstrated that the new criterion outperforms the commonly used criteria mentioned above. It was also shown that it selects the model order accurately and efficiently.; The developed analysing procedure has gone through extensive tests including numerical simulation tests and laboratory experiments. Impact tests on cantilever beams and frame, and low speed rolling element bearing with artificial faults experiments were conducted and the analysing technique was applied to detection and diagnosis. Not only was the procedure able to detect the faults, but an equation was also derived to quantitatively determine the severity of the faults based on the Prony parameters extracted from vibration signals. The procedure has proven to be an effective and efficient tool for Machine Condition Monitoring.; The application of wavelet transformation in condition monitoring was also explored with limited success.; Finally, recommendations and suggestions for future work were made with regard to Machine Condition Monitoring based on analysis of transient vibration signals. |
