Diagnosis And Prognosis Of Electrical And Mechanical Faults Using Wireless Sensor Networks And A Two-stage Neural Network Classifier

Diagnosis and isolation of electrical and mechanical problems in induction motors has always been a very challenging task. Some of the common problems in induction motors are: bearing, stator winding, and rotor bar failures. This thesis has three phases: The first one pertains to development of low-cost test-beds for simulating bearing faults and short circuit stator winding faults in a motor. Bearing fault is due to the failure of any of the components of the bearing and the stator winding fault is due to the failure of insulation between the windings. Bearing faults can be identified from the motor vibration signatures; where as the stator winding fault can be identified through the measurement of the fault voltage. Second, wireless modules for collection of voltage values and vibration data from the test-beds have been developed. Wireless sensors have been used because of their advantages over wired sensors in remote sensing and data collection without human intervention. Finally, a novel two-stage neural network is used to classify various bearing and short circuit faults. The first stage neural network estimates the principal components using the Generalized Hebbian Algorithm (GHA). Principal Component Analysis is used to reduce the dimensionality of the data and to extract the fault features. The second stage neural network uses a supervised learning vector quantization network (SLVQ) utilizing a self organizing map approach. This stage is used to classify various fault modes. This is followed by computation of performance metrics (Confusion Matrix, Receiver Operating Characteristics and Health Index) in order to determine the condition of the system at any instant of time and to predict the performance of the system in future. Neural networks have been used because of their flexibility in terms of online adaptive reformulation.