Fault diagnosis in induction motors using self-organizing neural networks and quantization error

In the paper an unsupervised neural network scheme for the identification and quantification of faults in induction motors is presented. Fault identification is performed using self-organizing maps neural networks. From the clusters generated by the neural network, the quantization error of each cluster is used to determine the fault magnitude. The general scheme is based on the motor's active and reactive instantaneous powers, in order to detect and diagnose faults whose characteristic frequencies are very close each other, such as broken rotor bars and oscillating loads. This network is trained using data obtained through the experimental measurements. Additional experimental data are later applied to the network in order to validate the proposal. Finally, it is demonstrated that the proposed strategy is able to correctly identifying and quantifying both faults, thus avoiding the need for an expert to perform the task.