Bearing signature extraction using an improved adaptive multiscale stochastic resonance method

Abstract

To improve the accuracy of bearing fault diagnosis, it is very important to extract weak features effectively from a noise signal. Mulitscale noise tuning stochastic resonance (MSTSR) has proved to be an extremely effective way for weak signal detection. However, the original MSTSR method, which is based on signal-to-noise ratio (SNR) index, requires some prior knowledge to select two key parameters (the cut-off wavelet decomposition level and the tuning parameter). To solve this problem, we present a modified kurtosis index for the MSTSR method in fault diagnosis of rolling element bearings. The proposed index can not only combine the advantages of kurtosis index and zero-crossing ratio but also enhance or weaken the impact degree of the zero-crossing ratio by correlation coefficient. Meanwhile, the adaptive approach gets rid of the requirement of any prior knowledge. The simulation and experimental results verify the proposed scheme is suitable for bearing fault diagnosis.