Jixie qiangdu (Jan 2020)
ROLLER BEARING FAULT RECOGNITION METHOD BASED ON CYCLIC AUTOCORRELATION AND MULTI-DOMAIN KERNEL LIMIT LEARNING MACHINE
Abstract
According to characteristics of the bearing signal,the second-order cyclic demodulation information was introduced into machine learning,and a multi-domain kernel extreme learning machine（MKELM） based on the combination of cyclic autocorrelation（CAF） frequency domain features and time domain features（TD） was proposed to accurately identify the bearing status.The algorithm constructed a CAF function based on the second-order cyclic characteristics of the bearing signal to extract the cyclic frequency domain features of the samples,then combined them with the time domain feature quantities of the samples.The matching factors of multi-domain feature vectors was designed to fuse TD and CAF feature vectors; finally,the fused CAF-TD sample features was input into the kernel extreme learning machine for cluster regression.The spindle bearing experimental results show that the cyclic frequency domain statistics extracted based on CAF can sensitively reflect the signal characteristics.Compared with the classic classifier,the CAF-TD multi-domain kernel extreme learning machine can extract more feature information from limited samples and obtain more accurate diagnostic result.
