Self-organizing map approach for classification of electricals rotor faults in induction motors

This paper presents electric motor rotor faults diagnosis using one kind of Artificial Neural Networks (ANN): Self Organizing maps (SOM) of Kohonen in clustering with two states of rotor: healthy rotor and faulty rotor. Major faults such as one broken rotor bars, two broken rotor bars and ring portion of the short circuit removed are considered. The SOM was trained using measurement data from stator currents. In this paper, two groups of parameters are used in the feature vector samples as inputs to neural networks. The groups are extracted from mathematical equation (1 ± 2.k. s)f_s with respectively the harmonic k=1 or k=2 in line current of motors with fault and healthy one. The effects of different network structures on the performance of the SOM are discussed. The results of the best map in this study show that the SOM gives satisfactory results and can in this case classify the type of motor fault where fault data from electric motors is available.