Model-based fault detection and diagnosis of BLDC motors working at variable speed using wavelet transform

Predictive Maintenance is becoming increasingly important in the automated industry, as maintaining equipment health is essential for the smooth flow of a manufacturing process. This thesis focuses on introducing a model-based technique for the predictive maintenance of robotic motors. The outcomes of this technique, when applied to practical scenarios as well as in simulations, are shown and discussed below.

The thesis aims to provide a strong theoretical justification to an idea, that was developed and tested successfully on real-life robot motor data. The idea involves applying the wavelet transform to the motor current and gathering the occurrence of frequencies relative to the central frequency (frequency ratio) to detect the presence of an anomalous frequency ratio. The enormous amount of information documented on experiments using Motor Current Signature Analysis (MCSA) for fault detection and diagnosis (FDD) is used here to diagnose the fault.

Simulations of two faulty models, Stator inter-turn winding short and eccentricity fault, have been developed and tested, at variable speeds, with this technique to provide assertive results. Also, the positive results obtained while the technique was applied to the robotic motor data has been presented and explained. This idea is different and more useful that MCSA alone as it works for variable speed conditions. Lastly, the thesis provides suggestions on how to expand on the technique to develop a useful predictive maintenance tool for robotic motors.