Abstract: Filter coefficient, time domain and frequency domain statistics are mainly used to describe the bearing performance degradation in bearing life prediction method based on data driven. Further studies show these features have neither sufficient sensitivity nor ascending or descending trend that lifetime prediction needs. The service lifetime of the same type of bearing varies considerably even on the same operating condition due to some random factor occurs in bearing manufacturing and working process. Therefore, searching for stable and effective features to describe the instant degradation of bearing is very important. A feature extraction method of bearing performance degradation based on time-frequency image fusion is proposed. The time-frequency energy distribution of bearing vibration signal is extracted by smoothed pseudo Wigner-Ville distribution(SPWVD), and the bearing performance degradation is described using the statistical feature calculated by gray level co-occurrence matrix(GLCM). Also, image fusion is used for combining the two directions of bearing vibration signals to eliminate the influence of the random factor. The result reveals that this method is applicable in the feature extraction of bearing performance degradation using 2012 PHM Competition bearing data sets.

Key words: Gray level co-occurrence matrix, Image fusion, Performance degradation feature, Time-frequency analysis