Bearing Fault Detection via Autoregressive Stator Current Modeling

This paper proposes a method for detecting developing bearing faults via stator current. Current-based condition monitoring offers significant economic savings and implementation advantages over vibration-based techniques. This method begins by filtering the stator current to remove most of the significant frequency content unrelated to bearing faults. Afterwards, the filtered stator current is used to train an autoregressive signal model. This model is first trained while the bearings are healthy, and a baseline spectrum is computed. As bearing health degrades, the modeled spectrum deviates from its baseline value; the mean spectral deviation is then used as the fault index. This fault index is able to track changes in machine vibration due to developing bearing faults. Due to the initial filtering process, this method is robust to many influences including variations in supply voltage, cyclical load torque variations, and other (nonbearing) fault sources. Experimental results from ten different bearings are used to verify the proficiency of this method.