Monitoring the Health Status of Rolling Bearings Based on Adaptive mean-shift

Abstract

In order to monitor the structural health condition of rolling bearings in real time, an adaptive mean-shift algorithm was proposed to replace the fixed step size in the Mean Shift algorithm with adaptive variable step size. Wavelet packet energy entropy was used to extract the characteristics of rolling bearing vibration signals, and a series of centroids were clustered by adaptive mean-shift after dimensionality reduction. The centroid in normal state was used as the reference centroid to calculate the offsets of all centroids (including reference centroid) and reference centroid. Experiments show that the adaptive mean-shift algorithm converges quickly and has high accuracy. The centroid offset can be used to monitor the structural health effectively.