Non-stationary signal processing for bearing health monitoring

Signals generated by transient vibrations in rolling bearings due to structural defects are non-stationary in nature, and reflect upon the operation condition of the bearing. Consequently, effective processing of non-stationary signals is critical to bearing health monitoring. This paper presents a comparative study of four representative time-frequency analysis techniques commonly employed for non-stationary signal processing. The analytical framework of the short-time Fourier transform, wavelet transform, wavelet packet transform, and Hilbert-Huang transform are first presented. The effectiveness of each technique in detecting transient features from a time-varying signal is then examined, using an analytically formulated test signal. Subsequently, the performance of each technique is experimentally evaluated, using realistic vibration signals measured from a bearing test system. The results demonstrate that selecting appropriate signal processing technique can significantly affect defect identification and consequently, improve the reliability of bearing health monitoring.