Detection of short circuit faults in 3-phase converter-fed induction motors using kernel SOMs

In this work we report the results of a comprehensive study involving the application of kernel self-organizing maps (KSOM) for early detection of interturn short-circuit faults in a three-phase converter-fed induction motor. For this purpose, two paradigms for developing KSOM-based classifiers are evaluated on the problem of interest, namely the gradient descent based KSOM (GD-KSOM) and the energy function based KSOM (EF-KSOM). Their performances are contrasted on a real-world dataset generated by means of a laboratory scale testbed that allows the simulation of different levels of interturn short-circuits (high and low impedance) for different load conditions. Feature vectors are built from the FFT-based spectrum analysis of the stator current, a non-invasive method known as the stator current signature. The performances of the aforementioned KSOM paradigms are evaluated for different kernel functions and for different neuron labeling strategies. The obtained results are compared with those achieved by standard SOM-based classifier.