Rotor bars fault detection by DFT spectral analysis and Extreme Learning Machine

Finding out about the damage of the induction motor particularly in the rotor bar is essential to ensure the continuity of the manufacturing process and to reduce the high cost of breakdown maintenance. In the past decades, the difficulty level to find fault in rotor bars has attracted many researcher to develop the solving method. Therefore, this work presents an alternative to discover fault on rotor bars based on analysis of stator currents (MCSA) using Discrete Fourier Transformation (DFT) and Extreme Learning Machine (ELM). In this study, ELM applied to detect abnormalities and fault of rotor bars on induction motor with no load condition. The performance of proposed algorithm would compared with Constructive Back Propagation Neural Network (CBP-NN). The structure of both methods consist of a single hidden layer with 20 neurons that activated by a tangent sigmoid function. In the experiment, the data input for all algorithms came from the normalized of DFT outcome. The test result shows that ELM is faster compared with CBP-NN about 0.0467 seconds in terms of time training. Although the accuracy of their training has a wide error deviation but for the new data pattern recognizing or its variation, ELM provides the output prediction closer to the target validation than CBP-NN. Thus ELM can be used as a top priority because of the high level of time efficiency and accurate predictions.