| Online monitoring of induction motor health is of increasing interest, as the industrial processes that depend on these motors become more complex and as the performance to cost ratio of monitoring technology (e.g. sensors, microprocessors) continues to increase. Much effort has been directed towards developing methods that use conventional signal processing and pattern classification techniques. This thesis addresses the main issues of detecting electrical and mechanical faults using the information provided by current and vibration sensors, within a probabilistic framework. The faults studied in this work are stator voltage imbalances (electrical), rotor bar failures (electro-mechanical) and bearing outer race failure (mechanical). These failures are representative of almost all the failures occurring in induction machines. Voltage imbalance has been the longest studied problem. These arise either due to the faulty topology of the electrical system or stator winding faults. Rotor bar faults have accounted for about 10% of failures in induction machines, they are the most difficult type of failures to detect. Bearing failure accounts for 40% of all known failures. There is no direct instrumentation known to detect these kinds of failures.; The framework developed provides a common solution methodology for the detection of all these different faults. The methodology utilizes a combination of machine modeling concepts, along with wavelet, and symbolic dynamic analysis to ensure early detection. Additionally the sensor fusion technique developed, presents a probabilistic approach to the problem of bearing faults in induction motors. The method extends the D-Markov process to combine the information from both electrical and mechanical sensors. This provides accurate detection, with a low false alarm rate. The technique has been simulated using the magnetic equivalent circuit method and experimentally validated on 2-hp squirrel cage induction motors. |
