Skip to Main Content
In this paper, we attempted to classify the acceleration signals of different working states of bearings by using the wavelet-based fractal analysis. Considering the similarity of the power spectrums between bearing vibration signals and 1/f processes signals, the principles of wavelet-based fractal analysis for bearing fault diagnosis are explored. To verify the feasibility and practicability of the presented method, experiments were carried out. Vibration signals from the bearings of different working conditions were gathered by acceleration sensors. After pre-treatment, the acceleration signals on both vertical and horizontal direction were decomposed to ten detailed signals at different wavelet scales by using the discrete wavelet transform. The variances of detailed signals at scales 2 to 8 were calculated and then fractal features of the acceleration signals were estimated from the slope of the variance progression. The fractal dimensions were significantly different among the different working conditions of the bearings and showed a high reproducibility. The results suggest that the wavelet-based fractal analysis is effective for classifying the working conditions of bearings. Moreover, the results of repeated trials demonstrate that the fractal dimensions generated by wavelet-based fractal analysis are stable and accurate in practice.