Feature Extraction of Rolling Bearing Incipient Fault Using an Improved SCA-Based UBSS Method

Abstract

Incipient fault signal of rolling bearing tends to have weak amplitude and low signal-to-noise ratio. Bearing vibration signal collected by sensor is generally a mixture of several unknown source signals. It is difficult to extract incipient fault features from bearing vibration signals directly. The blind source separation (BSS) algorithm is used to restore source signals from the mixed signals. However, ordinary BSS algorithm requires that the number of sensors cannot be less than the number of source signals. To solve these problems, an improved underdetermined blind source separation (UBSS) method is proposed in this paper. The bearing vibration signals are preprocessed by short-time Fourier transform (STFT). Then the fuzzy C-means (FCM) clustering and the weighted minimal L1 norm method are respectively used to evaluate the mixing matrix and recover the separated signals. At last, the features of rolling bearing incipient fault are extracted from the recovered signals by envelope demodulation. Experimental results verify the effectiveness of the proposed method.