Detection and Diagnosis of Rolling Element Bearing Faults Using Time Encoded Signal Processing and Recognition

This thesis presents a systematic study of using TESPAR (Time Encoded Signal Processing and Recognition), which presently is in use as an effective tool for speech recognition and shows great advantages in computational demands and accuracy, to develop a new technique for rolling element bearing fault detection and diagnosis.

The fundamentals of rolling element bearings are presented in line with different failure modes and relevant monitoring methods in the time domain, the frequency domain, the envelope spectrum and the wavelet analysis. These reviews show that vibration measurements are a proven and widely accepted data source for bearing monitoring of machinery.

This research thus has focused on developing TESPAR based approaches using vibration signals which are generated from bearings under different severities of faults located at the outer race, the inner race and the roller element. It firstly examines the theoretical basis of TESPAR and examines the diagnosis performance with a number of different simulated signals, which confirms that TESPAR based methods are able to resolve different signals by using their statistics including S-matrix, A-matrix and epoch duration, which paves a frame work to process and interpolate the bearing signal.

With understandings of the insights of bearing vibrations and TESPAR approaches a signal processing framework is then suggested to analyse bearing vibration signals. It consists of a pre-processing step which removes possible noise in the signal, a TESPAR coding step which converts the signal into TESPAR representations-TESPAR streams, a feature calculation step, which produces different TESPAR statistic parameters, and finally a diagnosis step which applies common statistics to TESPAR statistic parameters to obtain required results.

The TESPAR solution proposed in this thesis shows that discrimination between different bearing signal waveforms has been implemented successfully. TESPAR S- and A-Matrices were constructed for the cases tested and used together with statistical correlation to differentiate between the types of faults. However, the severities of bearing faults have been identified using another TESPAR feature called the mean absolute magnitude value calculated using epoch durations.

The performance of the TESPAR approach was then evaluated against the envelope spectrum; this being the most common method for bearing condition monitoring that is conducted in two terms; the process complexity and diagnosis performance.

A major contribution of this research programme is the development of a method that can provide improved detection and diagnosis of bearing fault types and severity of faults seeded into roller bearings.