Fault detection, diagnosis and prognosis of rolling element bearings: Frequency domain methods and hidden Markov modeling

In this thesis, two frequency domain algorithms are developed for estimating the running speed (revolutions per second of the inner race) and the bearing defect frequencies of rolling element ball bearings from the Fast Fourier Transform (FFT) of the vibration data.; Further, a new hidden Markov modeling (HMM) based fault detection and diagnosis scheme is also developed. Feature matrices extracted from amplitude demodulated vibration signals from both normal and faulty bearings were used to train HMMs to represent various bearing conditions. Features were based on the reflection coefficients of the polynomial transfer function of an auto-regressive (AR) model of the vibration signals. Fault(s) were detected by comparing the probabilities of the feature matrices extracted from the vibration signals given the HMM trained for the normal case with a pre-set threshold. An existing fault was diagnosed through selecting the HMM with the highest probability. The scheme was also adapted to diagnose multiple bearing faults. In this adapted scheme, features were based on the selected node energies of a wavelet packet decomposition of the vibration signal. For each fault, a different set of nodes was chosen. The node set for a fault consisted of the nodes whose energies are affected by the presence of that fault and minimally affected by the presence of other faults. Both schemes were tested with experimental data collected from an accelerometer measuring the vibration from the drive end ball bearing of an induction motor (Reliance Electric 2HP IQPreAlert) driven mechanical system and have proven to be very accurate.; Finally, a new HMM based bearing prognosis scheme is also presented. In this scheme, the magnitude spectra of vibration signals were divided into several equally spaced bands and the mean energies of these bands were used as features. Based on the features extracted from normal bearing vibration signals, an HMM was trained to model the normal condition. The probabilities of this HMM were then used to detect any defects and assess their severity. Experimental data collected from a bearing accelerated life test was used to verify the efficacy of the new scheme.