As degradation progresses, the vibration signal generated by the machines changes and contains fault related features. Frequently the vibration transients caused by interactions between the machinery components burdened by surface faults, clearances etc., exhibit nonlinear properties. Due to their short time duration, the energy of transients is distributed across the frequency domain. Depending on the nature of the nonlinearity and the local resonances, the frequency domain distribution may have different forms and feature different patterns. The classical higher order spectra (HOS) capable of detecting the relations between multiple frequency components are an efficient tool for the detection of nonlinear vibration transients. However, classical HOS such as bi- coherence or tricoherence, are effective only in cases where the distribution of the frequency components follows the specific distribution patterns. This dissertation proposes the new higher-order wavelet spectral crosscorrelation based vibration signal processing technologies overcoming the limitations of classical HOS and significantly extending the diagnostic capabilities of higher- order vibration signal analysis.First, the constraints of well-established classical HOS are explained on the example of wavelet bicoherence (WB) and wavelet tri-coherence (WT). Subsequently, the novel wavelet spectral crosscorrelation of order 3 (WSC3) and order 4 (WSC4) are proposed as superior vibration signal processing technologies overcoming the limitations of the classical HOS. The theory behind the novel WSC is explained in detail and its application for fault detection and condition monitoring is investigated. The vibration signals originating from the gearbox subjected to endurance test are processed with classical HOS of order 3 and 4, i.e., WB and WT and with novel WSC3 and WSC4. The obtained results are subjected to in-depth analysis that proves the superiority of the novel WSC over the classical HOS.
| Date of Award | 14 Aug 2024 |
|---|
| Original language | English |
|---|
| Supervisor | Len Gelman (Main Supervisor) & Andrew Ball (Co-Supervisor) |
|---|