New Features from Fourier Spectrum for Induction Machine Broken Bar Detection Using Statistical Pattern Recognition

A fault diagnosis system based on pattern recognition techniques is developed for cage induction machine broken bar detection. The developed algorithm uses the stator current and, if available, motor speed as input data. Several features are extracted from the Frequency spectrum of the current signal derived using the Fast Fourier Transform. The relevance of the features for the purpose of fault detection is investigated and verified. Statistical pattern recognition techniques namely the Bayes Minimum Error classifier is applied to distinguish motors with broken rotor bars from healthy ones. A series of experiments using a three phase 3 hp cage induction machine performed in different load and fault conditions is used to provide data for training and then testing the classifier. Experimental results confirm the ability of the proposed algorithm for detection of broken bar faults. Key-Words: Fault diagnosis, Broken bar, Induction machine, Feature extraction, Pattern recognition