A Hybrid TSA-Neural Network Approach for Induction Motor Faults Diagnosis

Abstract

The failure of the broken rotor bar is one of the foremost reasons of breakdown in induction motor. So far, a variety of electrical-based techniques have been developed to diagnosis the condition of induction motor rotor fault. The main object of this work is to develop a new concept of detection and classification of broken rotor bars in an induction motor. The proposed approach is based on the Time Synchronous Averaging method (TSA) associated with a neural network technique. A literature review shows that the conventional fault diagnosis methods, such as the stator current analysis, have limitations in detection faults in no load scenarios In addition to “no decision” problems. The hybrid TSA-Neural network approach presents a better solution to this problem. The residual currents are used as inputs for training a feedforward multilayer neural network. The conceived diagnosis system was tested under a different number of broken rotor bars. The obtained results prove the effectiveness and accuracy in the fault detection and the classification of the correct number of broken rotor bars.