Advanced Signal Processing Techniques for Bearing Fault Detection in Induction Motors

Abstract

This paper focuses on bearing fault (BF) detection in induction motors based on a combination of advanced signal processing tools with motor current signature analysis (MCSA). The applied tools consist on two main techniques: the power spectral density (PSD) estimation for spectral analysis and wavelet techniques for time-frequency analysis. In order to identify the signatures related to bearing faults in the current spectrum, the Welch and the Multiple Signal Classification (MUSIC) PSD estimations are applied. Since the PSD techniques are not appropriate for non-stationary conditions, the Discrete Wavelet Transform (DWT), the Wavelet Packet Transform (WPT) and the Stationary WPT (SWPT) are used and compared. The efficiency of the proposed approaches is verified by several experiments corresponding to three types of bearing faults. The wavelet analysis, especially the SWPT, shows its ability to identify the BF signatures more accurately than other wavelet techniques regardless the load level.