Fault detection using autoregressive modelling techniques

The use of spectral analysis for fault detection and diagnostics in real-time has been conservative due to concerns over large processing requirements, especially when large sample sizes and high sampling frequencies are used. In this work, it is shown how such concerns can be allayed, to a large extent, by Autoregressive (AR) modelling, as the AR method has enhanced resolution capabilities compared to the Fast Fourier Transform (FFT) technique even when small sample sizes are used and requires a sampling rate just slightly above the Nyquist rate to give good parameter estimates. The use of a parametric method of AR modelling for fault diagnosis and prognosis is a relatively new concept in the field of condition monitoring.;In this thesis, a new methodology is proposed that combines AR modelling techniques and pole-related spectral decomposition for the detection of incipient single-point bearing defects for a vibration-based condition monitoring system. Vibration signals obtained from the ball bearings of a dry vacuum pump operating in normal and faulty conditions are used as the test signals and are modelled as time-variant AR series.;The position of the poles, which are the roots of the AR coefficient polynomial, vary for every frame of vibration data. It is a known fact that as defects such as spalls and cracks start to appear on the ball bearings, the amplitude of the vibrations of characteristic defect frequencies increases. This is seen as the poles moving closer to the unit circle as the severity of the defect increases. Simple statistical indicators such as the power and frequency of each bearing defect spectral component can be extracted from the residual and position of the AR poles. These indicators can be effectively used for fault classification to distinguish between the no-fault and defective cases as the difference between them is significant.