An unsupervised neural network fault discriminating system implementation for on-line condition monitoring and diagnostics of induction machines

A new method for on-line classification of induction machine faults was proposed. The proposed system included an unsupervised neural network and two expert systems. This system was designed to detect and categorize incipient machine failures autonomously. Stator current (single phase) were collected from machines with inflicted damages and were analyzed by the system. The first of the two expert systems extracted stator current features thus allowing spectral information compression while the second classified fault types when a failure condition had been detected by the neural network. The results have indicated that this combination of neural and knowledge systems an effective method of classifying incipient failures in a line-operated induction machine.