Rolling Bearing Vibration Signal Feature Extraction Based on SVD and Mathematical Morphological Fractal Dimension Spectrum

When the fractal dimension is employed as the feature of rolling bearing vibration signals, there are not enough characteristics and the signal's nonlinearity cannot be demonstrated satisfactorily. This paper proposed a method to extract features from a fractal dimensions spectrum based on SVD(Singular Value Decomposition) and mathematical morphology (SVDFDS). The SVDFDS first reconstructs the Hankel matrix from the signal and then performs SVD decomposition. Next, calculate the fractal dimension of the first component, then put the second component to this signal. Every time a component superimposes to the signal, calculate its mathematical morphological fractal dimensions. These dimensions are made up of a spectrum of fractal dimensions. When compared to EMD (Empirical Mode Decomposition) and box-counting fractal dimensions, the experiment results revealed that the approach retrieved more features with improved discrimination. As a result, the approach suggested in this article may be a superior solution for extracting features from rolling bearing vibration data.