Competitive learning applied to detect broken rotor bars in induction motors

This paper describes an automatic, load independent procedure for the detection of broken bars in squirrel cage induction machines based on the analysis of the space-vector of voltages induced in the stator windings after supply disconnection. In this condition, no current flows in the stator windings and the voltages measurable at its terminals are due to flux produced by rotor current. If there are some broken bars and the rotor symmetry is compromised, the voltages induced in the stator windings results distorted and some particular harmonics increase their amplitudes. The diagnostic technique is based on monitoring these voltage harmonics by analyzing the space vector of the voltages induced in the stator windings via short-time MUSIC (STMUSIC) time-frequency pseudo-representation. The output data of MUSIC processing are fed into an unsupervised self-organizing neural network (NN) with an ABCL training algorithm that is able to successfully discriminate between data measured on healthy and faulty motors.