Analysis of rolling element bearing faults in rotating machinery: Experiments, modeling, fault detection and diagnosis

One of the most common rotating machines in use today is the electric motor, and the most common cause of failure in electric motors is rolling element bearing failure. Current bearing fault detection methods do not adequately consider the physics of bearing failures, thus they do not usually perform well in actual practice as a result. Therefore the goal of this dissertation is to provide a physics based foundation for rolling element bearing fault analysis.; In order to achieve this goal, a detailed mechanical model of an electric motor is developed with particular emphasis on the bearings and the introduction of bearing faults. Experiments are conducted on the motor in which single faults are introduced onto bearing surfaces, and the resulting vibration of the motor system is measured. These vibration measurements are used to parameterize and validate the detailed model of the motor/bearing system. The model is a 29 degree-of-freedom, nonlinear, time variant representation of the motor/bearing system, including a treatment of the Hertzian contact forces between the bearing elements as well as including a flexible, user-definable method for simulating faults ranging from initial spalls to advanced surface damage. Subsequent to model parameterization and verification, multiple and distributed bearing faults (i.e. advanced surface damage) are explored computationally. The results of these system response predictions indicate that envelope analysis, an established bearing fault detection method based on the demodulation of the system's vibration response, may be employed in some circumstances to detect and diagnose single faults, multiple faults, and distributed faults. More importantly, the relative magnitudes of the defect frequencies and especially their harmonics in the envelope spectra are shown in specific cases to indicate the nature of a particular bearing fault. A two degree of freedom, linear, time invariant simplification of the detailed model is developed, and the performance of this model relative to the real vibration measurements indicates that the simplified model is a viable approximation of the detailed model for applications which require relatively short computer run times and do not require the increased accuracy which the detailed model provides.