Classification and sensitivity analysis to detect fault in induction motors using an MLP network

This work is an investigation about the use of a Multilayer Perceptron Artificial Neural Network (MLP ANN) to detect stator winding short-circuit faults in a converter-fed induction motor. The algorithm uses six frequency components from the current spectrum as input variables. The data (samples) was acquired varying: (1) the frequencies imposed by inverter drive, (2) the load level, and (3) the fault extension of the induction motor. This articles approach is to investigate the influence of several aspects related to fault emulation and attributes selection over the categorization capacity of the classifier. Several hypotheses about those aforementioned influences are raised and analyzed. At the end, a classifier capable to identify the fault evolution is proposed and evaluated.