Enhanced information extraction from noisy vibration data for machinery fault detection and diagnosis

As key mechanical components, bearings and gearboxes are employed in most machines. To maintain efficient and safe operations in modern industries, their condition monitoring has received massive attention in recent years. This thesis focuses on the improvement of signal processing approaches to enhance the performance of vibration based monitoring techniques taking into account various data mechanisms and their associated periodic, impulsive, modulating, nonlinear coupling characteristics along with noise contamination. Through in-depth modelling, extensive simulations and experimental verifications upon different and combined faults that often occur in the bearings and gears of representative industrial gearbox systems, the thesis has made following main conclusions in acquiring accurate diagnostic information based on improved signal processing techniques:

1) Among a wide range of advanced approaches investigated, such as adaptive line enhancer (ALE), wavelet transforms, time synchronous averaging (TSA), Kurtogram analysis, and bispectrum representations, the modulation signal bispectrum based sideband estimator (MSB-SE) is regarded as the most powerful tool to enhance the periodic fault signatures as it has the unique property of simultaneous demodulation and noise reduction along with ease of implementation.

2) The proposed MSB-SE based robust detector can achieve optimal band selection and envelope spectrum analysis simultaneously and show more reliable results for bearing fault detection and diagnosis, compared with the popular Kurtogram analysis which highlights too much on localised impulses.

3) The proposed residual sideband analysis yields accurate and consistent diagnostic results of planetary gearboxes across wide operating conditions. This is because that the residual sidebands are much less influenced by inherent gear errors and can be enhanced by MSB analysis.

4) Combined faults in bearings and gears can be detected and separated by MSB analysis. To make the results more reliable, multiple slices of MSB-SE can be averaged to minimise redundant interferences and improve the diagnostic performance.