Neural Approach to Fault Detection in Three-phase Induction Motors

Three-phase induction motors play a key role in electromotive force production, although their widespread use in industrial applications leads to a variety of defects that can impair their normal operation. This paper proposes an alternative method, based on artificial neural networks, for classifying and detecting bearing faults in three-phase induction motors connected directly to the power grid. Experimental tests conducted in the laboratory consider practical conditions such as voltage unbalances and differing load torques. Analyses are performed in the time domain, and are based on electric motor quantities such as voltages and currents, which are acquired considering a half-cycle of the voltage grid. The performance and efficacy of the proposed fault detection method is evaluated and validated by using a personal computer to conduct online experimental tests, and by embedding in digital signal processors. The proposed method is also tested in electrical machines used in the sugar cane industry.