| Electric motors are at the core of engineering processes in industry. Although reliable, failure does occur and this can be costly to the operator in terms of lost production, inconvenience downtime and also in terms of the effect on safety implications. The implementation of effective maintenance can aid the asset management of these machines. This thesis addresses the development of condition monitoring of electric motors to enable appropriate adjustment of machine operation, downtime to be planned and spare parts to be ordered. The research was focused on condition monitoring of two types of electric motors: conventional three-phase induction motors, and a novel Transverse Flux Motor being developed by Rolls-Royce pic. A variety of seeded faults (phase imbalance, broken rotor bars, misalignment and failed insulation) in an induction motor could each be detected, diagnosed and severity assessed by unique changes within frequency spectra of measured data. In addition to the use of vibration and current, a main achievement was the demonstration of the capability of less-conventionally used measurement parameters (ie flux, acoustics, torque and speed). A unique approach to implementing and combining auto-regressive models, principal component analysis and confidence intervals, for providing a visually simple means for motor fault detection was also successfully developed. Magnetic flux measurement was implemented for condition monitoring of the Transverse Flux Motor. This work was original because of the way in which search coils were strategically placed to provide an indication of all perceived fault conditions. Importantly, an incipient demagnetisation fault was detected, located and severity assessed. In addition, valuable feedback was provided to the Transverse Flux Motor design team at Rolls-Royce. Determination of the alignment between components within this complex machine was of particular importance during this process. Collected data from the induction motor and Transverse Flux Motor also enabled a novel neural network (called Componential Coding) to be validated for practical machine monitoring. This computationally efficient network was advantageous for the simultaneous and automated analysis of multiple data channels. |
