Bearing fault detection in induction motors using the ROCOF and k-nearest neighbors algorithm

The actual tendency in electric systems for industrial and academic applications is to lead them to the smart systems philosophy. Nowadays, electronic instrumentation is an interesting area for researchers for electric motors’ control, monitoring, and maintenance purposes. Motor current signal analysis is one of the most used methods for fault detection in electric machines due to its noninvasive nature and the vast amount of information it contains; several techniques have been developed for bearing fault detection based on current signals, commonly through a spectrum analysis. Because of this, this paper proposes a methodology based on the rate of change of frequency estimation and a machine learning classifier (k-nearest neighbors). In this work, bearing ball damage and outer race damage scenarios were analyzed using the proposed method via the Digital Taylor-Fourier Transform, resulting in accuracy percentages of 97% and 94% for each case, respectively. That methodology does not require characterization supervised by a qualified person.