State Assessment of Rolling Bearings Based on the Multiscale Bubble Entropy

Abstract

An improved bubble entropy called multiscale bubble entropy (MBE) is proposed based on the multiscale processing, and then the application of MBE in bearing state assessment is investigated. Firstly, the MBE features are extracted from the collected vibration signals of the bearing with the whole life, and then dimension reduction is performed with principal component analysis (PCA). Secondly, a performance degradation indicator (PDI) based on the first smoothed principal component is constructed to represent the bearing condition monitoring. In the following, the fault type of bearings is identified with the principal components of features from different fault types and support vector machine with directed acyclic graph (DAG-SVM). Two groups of experimental data are investigated to illustrate the availability of the proposed feature in bearing condition monitoring and fault diagnosis. The results show that the trend of PDI has good monotonicity to represent the condition monitoring of the bearing, while the accuracy of fault classification is high and stable.